1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // SSE 1 & 2 - Move Instructions
120 //===----------------------------------------------------------------------===//
122 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
123 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
124 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
126 // Loading from memory automatically zeroing upper bits.
127 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
128 PatFrag mem_pat, string OpcodeStr> :
129 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
131 [(set RC:$dst, (mem_pat addr:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
139 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
140 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
141 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
143 let canFoldAsLoad = 1, isReMaterializable = 1 in {
144 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
146 let AddedComplexity = 20 in
147 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
150 let Constraints = "$src1 = $dst" in {
151 def MOVSSrr : sse12_move_rr<FR32, v4f32,
152 "movss\t{$src2, $dst|$dst, $src2}">, XS;
153 def MOVSDrr : sse12_move_rr<FR64, v2f64,
154 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
157 let canFoldAsLoad = 1, isReMaterializable = 1 in {
158 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
160 let AddedComplexity = 20 in
161 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
164 let AddedComplexity = 15 in {
165 // Extract the low 32-bit value from one vector and insert it into another.
166 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
167 (MOVSSrr (v4f32 VR128:$src1),
168 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
169 // Extract the low 64-bit value from one vector and insert it into another.
170 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
171 (MOVSDrr (v2f64 VR128:$src1),
172 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
175 // Implicitly promote a 32-bit scalar to a vector.
176 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
177 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
178 // Implicitly promote a 64-bit scalar to a vector.
179 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
180 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
181 // Implicitly promote a 32-bit scalar to a vector.
182 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
183 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
184 // Implicitly promote a 64-bit scalar to a vector.
185 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
186 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
188 let AddedComplexity = 20 in {
189 // MOVSSrm zeros the high parts of the register; represent this
190 // with SUBREG_TO_REG.
191 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
192 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
193 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 // MOVSDrm zeros the high parts of the register; represent this
198 // with SUBREG_TO_REG.
199 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
200 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
201 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzload addr:$src)),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
211 // Store scalar value to memory.
212 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
213 "movss\t{$src, $dst|$dst, $src}",
214 [(store FR32:$src, addr:$dst)]>;
215 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
216 "movsd\t{$src, $dst|$dst, $src}",
217 [(store FR64:$src, addr:$dst)]>;
219 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
220 "movss\t{$src, $dst|$dst, $src}",
221 [(store FR32:$src, addr:$dst)]>, XS, VEX;
222 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
223 "movsd\t{$src, $dst|$dst, $src}",
224 [(store FR64:$src, addr:$dst)]>, XD, VEX;
226 // Extract and store.
227 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
230 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
231 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
236 // Move Aligned/Unaligned floating point values
237 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
238 X86MemOperand x86memop, PatFrag ld_frag,
239 string asm, Domain d,
240 bit IsReMaterializable = 1> {
241 let neverHasSideEffects = 1 in
242 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
243 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
244 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
245 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
246 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
247 [(set RC:$dst, (ld_frag addr:$src))], d>;
250 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
251 "movaps", SSEPackedSingle>, VEX;
252 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
253 "movapd", SSEPackedDouble>, OpSize, VEX;
254 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
255 "movups", SSEPackedSingle>, VEX;
256 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
257 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
259 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
260 "movaps", SSEPackedSingle>, VEX;
261 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
262 "movapd", SSEPackedDouble>, OpSize, VEX;
263 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
264 "movups", SSEPackedSingle>, VEX;
265 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
266 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
267 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
268 "movaps", SSEPackedSingle>, TB;
269 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
270 "movapd", SSEPackedDouble>, TB, OpSize;
271 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
272 "movups", SSEPackedSingle>, TB;
273 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
274 "movupd", SSEPackedDouble, 0>, TB, OpSize;
276 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
277 "movaps\t{$src, $dst|$dst, $src}",
278 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
279 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
280 "movapd\t{$src, $dst|$dst, $src}",
281 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
282 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
283 "movups\t{$src, $dst|$dst, $src}",
284 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
285 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
286 "movupd\t{$src, $dst|$dst, $src}",
287 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
288 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
289 "movaps\t{$src, $dst|$dst, $src}",
290 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
291 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
292 "movapd\t{$src, $dst|$dst, $src}",
293 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
294 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
295 "movups\t{$src, $dst|$dst, $src}",
296 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
297 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
298 "movupd\t{$src, $dst|$dst, $src}",
299 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
302 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
303 (VMOVUPSYmr addr:$dst, VR256:$src)>;
305 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
306 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
307 (VMOVUPDYmr addr:$dst, VR256:$src)>;
309 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
310 "movaps\t{$src, $dst|$dst, $src}",
311 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
312 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
313 "movapd\t{$src, $dst|$dst, $src}",
314 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
315 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
316 "movups\t{$src, $dst|$dst, $src}",
317 [(store (v4f32 VR128:$src), addr:$dst)]>;
318 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movupd\t{$src, $dst|$dst, $src}",
320 [(store (v2f64 VR128:$src), addr:$dst)]>;
322 // Intrinsic forms of MOVUPS/D load and store
323 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
324 (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
327 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
328 (ins f128mem:$dst, VR128:$src),
329 "movupd\t{$src, $dst|$dst, $src}",
330 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
332 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
333 "movups\t{$src, $dst|$dst, $src}",
334 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
335 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
336 "movupd\t{$src, $dst|$dst, $src}",
337 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
339 // Move Low/High packed floating point values
340 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
341 PatFrag mov_frag, string base_opc,
343 def PSrm : PI<opc, MRMSrcMem,
344 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
345 !strconcat(base_opc, "s", asm_opr),
348 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
349 SSEPackedSingle>, TB;
351 def PDrm : PI<opc, MRMSrcMem,
352 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
353 !strconcat(base_opc, "d", asm_opr),
354 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
355 (scalar_to_vector (loadf64 addr:$src2)))))],
356 SSEPackedDouble>, TB, OpSize;
359 let AddedComplexity = 20 in {
360 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
361 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
362 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
363 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
365 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
366 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
367 "\t{$src2, $dst|$dst, $src2}">;
368 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
369 "\t{$src2, $dst|$dst, $src2}">;
372 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
373 "movlps\t{$src, $dst|$dst, $src}",
374 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
375 (iPTR 0))), addr:$dst)]>, VEX;
376 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
377 "movlpd\t{$src, $dst|$dst, $src}",
378 [(store (f64 (vector_extract (v2f64 VR128:$src),
379 (iPTR 0))), addr:$dst)]>, VEX;
380 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
381 "movlps\t{$src, $dst|$dst, $src}",
382 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
383 (iPTR 0))), addr:$dst)]>;
384 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
385 "movlpd\t{$src, $dst|$dst, $src}",
386 [(store (f64 (vector_extract (v2f64 VR128:$src),
387 (iPTR 0))), addr:$dst)]>;
389 // v2f64 extract element 1 is always custom lowered to unpack high to low
390 // and extract element 0 so the non-store version isn't too horrible.
391 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
392 "movhps\t{$src, $dst|$dst, $src}",
393 [(store (f64 (vector_extract
394 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
395 (undef)), (iPTR 0))), addr:$dst)]>,
397 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
398 "movhpd\t{$src, $dst|$dst, $src}",
399 [(store (f64 (vector_extract
400 (v2f64 (unpckh VR128:$src, (undef))),
401 (iPTR 0))), addr:$dst)]>,
403 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movhps\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract
406 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
407 (undef)), (iPTR 0))), addr:$dst)]>;
408 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movhpd\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract
411 (v2f64 (unpckh VR128:$src, (undef))),
412 (iPTR 0))), addr:$dst)]>;
414 let AddedComplexity = 20 in {
415 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
416 (ins VR128:$src1, VR128:$src2),
417 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
419 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
421 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
422 (ins VR128:$src1, VR128:$src2),
423 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
425 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
428 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
429 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
430 (ins VR128:$src1, VR128:$src2),
431 "movlhps\t{$src2, $dst|$dst, $src2}",
433 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
434 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
435 (ins VR128:$src1, VR128:$src2),
436 "movhlps\t{$src2, $dst|$dst, $src2}",
438 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
441 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
442 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
443 let AddedComplexity = 20 in {
444 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
445 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
446 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
447 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
450 //===----------------------------------------------------------------------===//
451 // SSE 1 & 2 - Conversion Instructions
452 //===----------------------------------------------------------------------===//
454 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
455 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
457 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
458 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
460 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
463 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
464 X86MemOperand x86memop, string asm> {
465 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
467 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
471 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
472 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
473 string asm, Domain d> {
474 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
475 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
476 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
477 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
480 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
481 X86MemOperand x86memop, string asm> {
482 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
483 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
484 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
485 (ins DstRC:$src1, x86memop:$src),
486 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
489 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
490 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
491 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
492 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
494 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
495 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
496 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
497 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
500 // The assembler can recognize rr 64-bit instructions by seeing a rxx
501 // register, but the same isn't true when only using memory operands,
502 // provide other assembly "l" and "q" forms to address this explicitly
503 // where appropriate to do so.
504 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
506 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
508 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
510 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
512 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
515 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
516 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
517 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
518 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
519 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
520 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
521 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
522 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
523 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
524 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
525 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
526 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
527 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
528 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
529 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
530 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
532 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
533 // and/or XMM operand(s).
535 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
536 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
538 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
539 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
540 [(set DstRC:$dst, (Int SrcRC:$src))]>;
541 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
542 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
543 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
546 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
547 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
548 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
549 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
551 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
552 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
553 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
554 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
555 (ins DstRC:$src1, x86memop:$src2),
557 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
558 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
559 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
562 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
563 f32mem, load, "cvtss2si">, XS, VEX;
564 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
565 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
567 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
568 f128mem, load, "cvtsd2si">, XD, VEX;
569 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
570 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
573 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
574 // Get rid of this hack or rename the intrinsics, there are several
575 // intructions that only match with the intrinsic form, why create duplicates
576 // to let them be recognized by the assembler?
577 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
578 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
579 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
580 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
581 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
582 f32mem, load, "cvtss2si">, XS;
583 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
584 f32mem, load, "cvtss2si{q}">, XS, REX_W;
585 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
586 f128mem, load, "cvtsd2si{l}">, XD;
587 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
588 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
591 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
592 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
593 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
594 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
596 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
597 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
598 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
599 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
602 let Constraints = "$src1 = $dst" in {
603 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
604 int_x86_sse_cvtsi2ss, i32mem, loadi32,
606 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
607 int_x86_sse_cvtsi642ss, i64mem, loadi64,
608 "cvtsi2ss{q}">, XS, REX_W;
609 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
610 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
612 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
613 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
614 "cvtsi2sd">, XD, REX_W;
619 // Aliases for intrinsics
620 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
621 f32mem, load, "cvttss2si">, XS, VEX;
622 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
623 int_x86_sse_cvttss2si64, f32mem, load,
624 "cvttss2si">, XS, VEX, VEX_W;
625 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
626 f128mem, load, "cvttsd2si">, XD, VEX;
627 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
628 int_x86_sse2_cvttsd2si64, f128mem, load,
629 "cvttsd2si">, XD, VEX, VEX_W;
630 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
631 f32mem, load, "cvttss2si">, XS;
632 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
633 int_x86_sse_cvttss2si64, f32mem, load,
634 "cvttss2si{q}">, XS, REX_W;
635 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
636 f128mem, load, "cvttsd2si">, XD;
637 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
638 int_x86_sse2_cvttsd2si64, f128mem, load,
639 "cvttsd2si{q}">, XD, REX_W;
641 let Pattern = []<dag> in {
642 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
643 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
644 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
645 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
647 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
648 "cvtdq2ps\t{$src, $dst|$dst, $src}",
649 SSEPackedSingle>, TB, VEX;
650 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
651 "cvtdq2ps\t{$src, $dst|$dst, $src}",
652 SSEPackedSingle>, TB, VEX;
654 let Pattern = []<dag> in {
655 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
656 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
657 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
658 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
659 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
660 "cvtdq2ps\t{$src, $dst|$dst, $src}",
661 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
666 // Convert scalar double to scalar single
667 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
668 (ins FR64:$src1, FR64:$src2),
669 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
671 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
672 (ins FR64:$src1, f64mem:$src2),
673 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
674 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
675 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
678 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
679 "cvtsd2ss\t{$src, $dst|$dst, $src}",
680 [(set FR32:$dst, (fround FR64:$src))]>;
681 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
682 "cvtsd2ss\t{$src, $dst|$dst, $src}",
683 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
684 Requires<[HasSSE2, OptForSize]>;
686 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
687 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
689 let Constraints = "$src1 = $dst" in
690 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
691 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
693 // Convert scalar single to scalar double
694 // SSE2 instructions with XS prefix
695 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
696 (ins FR32:$src1, FR32:$src2),
697 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
698 []>, XS, Requires<[HasAVX]>, VEX_4V;
699 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
700 (ins FR32:$src1, f32mem:$src2),
701 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
702 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
703 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
706 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
707 "cvtss2sd\t{$src, $dst|$dst, $src}",
708 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
710 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
711 "cvtss2sd\t{$src, $dst|$dst, $src}",
712 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
713 Requires<[HasSSE2, OptForSize]>;
715 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
719 VR128:$src2))]>, XS, VEX_4V,
721 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
722 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
725 (load addr:$src2)))]>, XS, VEX_4V,
727 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
728 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
729 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
730 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
731 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
734 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
735 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
738 (load addr:$src2)))]>, XS,
742 def : Pat<(extloadf32 addr:$src),
743 (CVTSS2SDrr (MOVSSrm addr:$src))>,
744 Requires<[HasSSE2, OptForSpeed]>;
746 // Convert doubleword to packed single/double fp
747 // SSE2 instructions without OpSize prefix
748 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
749 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
750 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
751 TB, VEX, Requires<[HasAVX]>;
752 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
753 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
754 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
755 (bitconvert (memopv2i64 addr:$src))))]>,
756 TB, VEX, Requires<[HasAVX]>;
757 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
758 "cvtdq2ps\t{$src, $dst|$dst, $src}",
759 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
760 TB, Requires<[HasSSE2]>;
761 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
762 "cvtdq2ps\t{$src, $dst|$dst, $src}",
763 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
764 (bitconvert (memopv2i64 addr:$src))))]>,
765 TB, Requires<[HasSSE2]>;
767 // FIXME: why the non-intrinsic version is described as SSE3?
768 // SSE2 instructions with XS prefix
769 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
770 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
771 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
772 XS, VEX, Requires<[HasAVX]>;
773 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
774 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
775 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
776 (bitconvert (memopv2i64 addr:$src))))]>,
777 XS, VEX, Requires<[HasAVX]>;
778 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
779 "cvtdq2pd\t{$src, $dst|$dst, $src}",
780 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
781 XS, Requires<[HasSSE2]>;
782 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
783 "cvtdq2pd\t{$src, $dst|$dst, $src}",
784 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
785 (bitconvert (memopv2i64 addr:$src))))]>,
786 XS, Requires<[HasSSE2]>;
789 // Convert packed single/double fp to doubleword
790 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
791 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
792 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
794 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
796 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
798 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
800 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
803 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
804 "cvtps2dq\t{$src, $dst|$dst, $src}",
805 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
807 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
809 "cvtps2dq\t{$src, $dst|$dst, $src}",
810 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
811 (memop addr:$src)))]>, VEX;
812 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
813 "cvtps2dq\t{$src, $dst|$dst, $src}",
814 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
815 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
816 "cvtps2dq\t{$src, $dst|$dst, $src}",
817 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
818 (memop addr:$src)))]>;
820 // SSE2 packed instructions with XD prefix
821 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
822 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
823 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
824 XD, VEX, Requires<[HasAVX]>;
825 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
826 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
827 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
828 (memop addr:$src)))]>,
829 XD, VEX, Requires<[HasAVX]>;
830 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
831 "cvtpd2dq\t{$src, $dst|$dst, $src}",
832 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
833 XD, Requires<[HasSSE2]>;
834 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
835 "cvtpd2dq\t{$src, $dst|$dst, $src}",
836 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
837 (memop addr:$src)))]>,
838 XD, Requires<[HasSSE2]>;
841 // Convert with truncation packed single/double fp to doubleword
842 // SSE2 packed instructions with XS prefix
843 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
844 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
845 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
847 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
848 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
849 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
850 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
851 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
852 "cvttps2dq\t{$src, $dst|$dst, $src}",
854 (int_x86_sse2_cvttps2dq VR128:$src))]>;
855 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
856 "cvttps2dq\t{$src, $dst|$dst, $src}",
858 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
861 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
862 "vcvttps2dq\t{$src, $dst|$dst, $src}",
864 (int_x86_sse2_cvttps2dq VR128:$src))]>,
865 XS, VEX, Requires<[HasAVX]>;
866 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
867 "vcvttps2dq\t{$src, $dst|$dst, $src}",
868 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
869 (memop addr:$src)))]>,
870 XS, VEX, Requires<[HasAVX]>;
872 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
874 "cvttpd2dq\t{$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
877 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
879 "cvttpd2dq\t{$src, $dst|$dst, $src}",
880 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
881 (memop addr:$src)))]>, VEX;
882 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
883 "cvttpd2dq\t{$src, $dst|$dst, $src}",
884 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
885 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
886 "cvttpd2dq\t{$src, $dst|$dst, $src}",
887 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
888 (memop addr:$src)))]>;
890 // The assembler can recognize rr 256-bit instructions by seeing a ymm
891 // register, but the same isn't true when using memory operands instead.
892 // Provide other assembly rr and rm forms to address this explicitly.
893 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
894 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
895 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
896 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
899 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
900 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
901 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
902 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
905 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
906 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
907 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
908 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
910 // Convert packed single to packed double
911 let Predicates = [HasAVX] in {
912 // SSE2 instructions without OpSize prefix
913 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
914 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
915 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
916 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
917 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
918 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
919 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
920 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
922 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
923 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
924 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
925 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
927 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
928 "vcvtps2pd\t{$src, $dst|$dst, $src}",
929 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
930 VEX, Requires<[HasAVX]>;
931 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
932 "vcvtps2pd\t{$src, $dst|$dst, $src}",
933 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
934 (load addr:$src)))]>,
935 VEX, Requires<[HasAVX]>;
936 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
937 "cvtps2pd\t{$src, $dst|$dst, $src}",
938 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
939 TB, Requires<[HasSSE2]>;
940 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
941 "cvtps2pd\t{$src, $dst|$dst, $src}",
942 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
943 (load addr:$src)))]>,
944 TB, Requires<[HasSSE2]>;
946 // Convert packed double to packed single
947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
948 // register, but the same isn't true when using memory operands instead.
949 // Provide other assembly rr and rm forms to address this explicitly.
950 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
952 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
953 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
963 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
964 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
965 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
966 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
967 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
968 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
969 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
972 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
973 "cvtpd2ps\t{$src, $dst|$dst, $src}",
974 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
975 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
977 "cvtpd2ps\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
979 (memop addr:$src)))]>;
980 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
981 "cvtpd2ps\t{$src, $dst|$dst, $src}",
982 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
983 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
984 "cvtpd2ps\t{$src, $dst|$dst, $src}",
985 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
986 (memop addr:$src)))]>;
988 // AVX 256-bit register conversion intrinsics
989 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
990 // whenever possible to avoid declaring two versions of each one.
991 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
992 (VCVTDQ2PSYrr VR256:$src)>;
993 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
994 (VCVTDQ2PSYrm addr:$src)>;
996 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
997 (VCVTPD2PSYrr VR256:$src)>;
998 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
999 (VCVTPD2PSYrm addr:$src)>;
1001 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1002 (VCVTPS2DQYrr VR256:$src)>;
1003 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1004 (VCVTPS2DQYrm addr:$src)>;
1006 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1007 (VCVTPS2PDYrr VR128:$src)>;
1008 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1009 (VCVTPS2PDYrm addr:$src)>;
1011 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1012 (VCVTTPD2DQYrr VR256:$src)>;
1013 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1014 (VCVTTPD2DQYrm addr:$src)>;
1016 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1017 (VCVTTPS2DQYrr VR256:$src)>;
1018 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1019 (VCVTTPS2DQYrm addr:$src)>;
1021 //===----------------------------------------------------------------------===//
1022 // SSE 1 & 2 - Compare Instructions
1023 //===----------------------------------------------------------------------===//
1025 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1026 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1027 string asm, string asm_alt> {
1028 let isAsmParserOnly = 1 in {
1029 def rr : SIi8<0xC2, MRMSrcReg,
1030 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1033 def rm : SIi8<0xC2, MRMSrcMem,
1034 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1038 // Accept explicit immediate argument form instead of comparison code.
1039 def rr_alt : SIi8<0xC2, MRMSrcReg,
1040 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1043 def rm_alt : SIi8<0xC2, MRMSrcMem,
1044 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1048 let neverHasSideEffects = 1 in {
1049 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1050 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1051 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1053 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1054 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1055 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1059 let Constraints = "$src1 = $dst" in {
1060 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1061 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1062 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1063 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1064 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1065 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1066 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1067 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1068 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1069 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1070 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1071 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1072 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1073 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1074 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1075 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1077 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1078 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1079 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1080 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1081 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1082 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1083 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1084 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1085 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1086 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1087 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1088 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1089 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1092 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1093 Intrinsic Int, string asm> {
1094 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1095 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1096 [(set VR128:$dst, (Int VR128:$src1,
1097 VR128:$src, imm:$cc))]>;
1098 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1099 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1100 [(set VR128:$dst, (Int VR128:$src1,
1101 (load addr:$src), imm:$cc))]>;
1104 // Aliases to match intrinsics which expect XMM operand(s).
1105 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1106 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1108 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1109 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1111 let Constraints = "$src1 = $dst" in {
1112 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1113 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1114 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1115 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1119 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1120 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1121 ValueType vt, X86MemOperand x86memop,
1122 PatFrag ld_frag, string OpcodeStr, Domain d> {
1123 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1124 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1125 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1126 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1127 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1128 [(set EFLAGS, (OpNode (vt RC:$src1),
1129 (ld_frag addr:$src2)))], d>;
1132 let Defs = [EFLAGS] in {
1133 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1134 "ucomiss", SSEPackedSingle>, VEX;
1135 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1136 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1137 let Pattern = []<dag> in {
1138 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1139 "comiss", SSEPackedSingle>, VEX;
1140 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1141 "comisd", SSEPackedDouble>, OpSize, VEX;
1144 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1145 load, "ucomiss", SSEPackedSingle>, VEX;
1146 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1147 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1149 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1150 load, "comiss", SSEPackedSingle>, VEX;
1151 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1152 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1153 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1154 "ucomiss", SSEPackedSingle>, TB;
1155 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1156 "ucomisd", SSEPackedDouble>, TB, OpSize;
1158 let Pattern = []<dag> in {
1159 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1160 "comiss", SSEPackedSingle>, TB;
1161 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1162 "comisd", SSEPackedDouble>, TB, OpSize;
1165 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1166 load, "ucomiss", SSEPackedSingle>, TB;
1167 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1168 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1170 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1171 "comiss", SSEPackedSingle>, TB;
1172 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1173 "comisd", SSEPackedDouble>, TB, OpSize;
1174 } // Defs = [EFLAGS]
1176 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1177 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1178 Intrinsic Int, string asm, string asm_alt,
1180 let isAsmParserOnly = 1 in {
1181 def rri : PIi8<0xC2, MRMSrcReg,
1182 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1183 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1184 def rmi : PIi8<0xC2, MRMSrcMem,
1185 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1186 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1189 // Accept explicit immediate argument form instead of comparison code.
1190 def rri_alt : PIi8<0xC2, MRMSrcReg,
1191 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1193 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1194 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1198 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1199 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1200 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1201 SSEPackedSingle>, VEX_4V;
1202 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1203 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1204 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1205 SSEPackedDouble>, OpSize, VEX_4V;
1206 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1207 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1208 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1209 SSEPackedSingle>, VEX_4V;
1210 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1211 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1212 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1213 SSEPackedDouble>, OpSize, VEX_4V;
1214 let Constraints = "$src1 = $dst" in {
1215 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1216 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1217 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1218 SSEPackedSingle>, TB;
1219 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1220 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1221 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1222 SSEPackedDouble>, TB, OpSize;
1225 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1226 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1227 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1228 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1229 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1230 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1231 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1232 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1234 //===----------------------------------------------------------------------===//
1235 // SSE 1 & 2 - Shuffle Instructions
1236 //===----------------------------------------------------------------------===//
1238 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1239 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1240 ValueType vt, string asm, PatFrag mem_frag,
1241 Domain d, bit IsConvertibleToThreeAddress = 0> {
1242 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1243 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1244 [(set RC:$dst, (vt (shufp:$src3
1245 RC:$src1, (mem_frag addr:$src2))))], d>;
1246 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1247 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1248 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1250 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1253 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1254 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1255 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1256 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1257 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1258 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1259 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1260 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1261 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1262 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1263 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1264 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1266 let Constraints = "$src1 = $dst" in {
1267 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1268 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1269 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1271 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1272 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1273 memopv2f64, SSEPackedDouble>, TB, OpSize;
1276 //===----------------------------------------------------------------------===//
1277 // SSE 1 & 2 - Unpack Instructions
1278 //===----------------------------------------------------------------------===//
1280 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1281 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1282 PatFrag mem_frag, RegisterClass RC,
1283 X86MemOperand x86memop, string asm,
1285 def rr : PI<opc, MRMSrcReg,
1286 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1288 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1289 def rm : PI<opc, MRMSrcMem,
1290 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1292 (vt (OpNode RC:$src1,
1293 (mem_frag addr:$src2))))], d>;
1296 let AddedComplexity = 10 in {
1297 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1298 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1299 SSEPackedSingle>, VEX_4V;
1300 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1301 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1302 SSEPackedDouble>, OpSize, VEX_4V;
1303 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1304 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1305 SSEPackedSingle>, VEX_4V;
1306 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1307 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1308 SSEPackedDouble>, OpSize, VEX_4V;
1310 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1311 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1312 SSEPackedSingle>, VEX_4V;
1313 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1314 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1315 SSEPackedDouble>, OpSize, VEX_4V;
1316 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1317 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1318 SSEPackedSingle>, VEX_4V;
1319 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1320 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1321 SSEPackedDouble>, OpSize, VEX_4V;
1323 let Constraints = "$src1 = $dst" in {
1324 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1325 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1326 SSEPackedSingle>, TB;
1327 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1328 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1329 SSEPackedDouble>, TB, OpSize;
1330 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1331 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1332 SSEPackedSingle>, TB;
1333 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1334 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1335 SSEPackedDouble>, TB, OpSize;
1336 } // Constraints = "$src1 = $dst"
1337 } // AddedComplexity
1339 //===----------------------------------------------------------------------===//
1340 // SSE 1 & 2 - Extract Floating-Point Sign mask
1341 //===----------------------------------------------------------------------===//
1343 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1344 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1346 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1347 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1348 [(set GR32:$dst, (Int RC:$src))], d>;
1349 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1350 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1354 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1355 "movmskps", SSEPackedSingle>, VEX;
1356 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1357 "movmskpd", SSEPackedDouble>, OpSize,
1359 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1360 "movmskps", SSEPackedSingle>, VEX;
1361 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1362 "movmskpd", SSEPackedDouble>, OpSize,
1364 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1365 SSEPackedSingle>, TB;
1366 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1367 SSEPackedDouble>, TB, OpSize;
1370 def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
1371 "movmskpd\t{$src, $dst|$dst, $src}",
1372 [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1373 def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1374 "movmskpd\t{$src, $dst|$dst, $src}",
1375 [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1376 def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
1377 "movmskps\t{$src, $dst|$dst, $src}",
1378 [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1379 def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1380 "movmskps\t{$src, $dst|$dst, $src}",
1381 [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1384 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1385 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1386 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1387 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1389 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1390 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1391 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1392 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1395 //===----------------------------------------------------------------------===//
1396 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1397 //===----------------------------------------------------------------------===//
1399 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1400 // names that start with 'Fs'.
1402 // Alias instructions that map fld0 to pxor for sse.
1403 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1404 canFoldAsLoad = 1 in {
1405 // FIXME: Set encoding to pseudo!
1406 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1407 [(set FR32:$dst, fp32imm0)]>,
1408 Requires<[HasSSE1]>, TB, OpSize;
1409 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1410 [(set FR64:$dst, fpimm0)]>,
1411 Requires<[HasSSE2]>, TB, OpSize;
1412 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1413 [(set FR32:$dst, fp32imm0)]>,
1414 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1415 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1416 [(set FR64:$dst, fpimm0)]>,
1417 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1420 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1421 // bits are disregarded.
1422 let neverHasSideEffects = 1 in {
1423 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1424 "movaps\t{$src, $dst|$dst, $src}", []>;
1425 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1426 "movapd\t{$src, $dst|$dst, $src}", []>;
1429 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1430 // bits are disregarded.
1431 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1432 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1433 "movaps\t{$src, $dst|$dst, $src}",
1434 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1435 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1436 "movapd\t{$src, $dst|$dst, $src}",
1437 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1440 //===----------------------------------------------------------------------===//
1441 // SSE 1 & 2 - Logical Instructions
1442 //===----------------------------------------------------------------------===//
1444 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1446 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1448 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1449 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1451 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1452 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1454 let Constraints = "$src1 = $dst" in {
1455 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1456 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1458 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1459 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1463 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1464 let mayLoad = 0 in {
1465 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1466 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1467 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1470 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1471 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1473 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1475 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1477 let Pattern = []<dag> in {
1478 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1479 !strconcat(OpcodeStr, "ps"), f128mem,
1480 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1481 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1482 (memopv2i64 addr:$src2)))], 0>, VEX_4V;
1484 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1485 !strconcat(OpcodeStr, "pd"), f128mem,
1486 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1487 (bc_v2i64 (v2f64 VR128:$src2))))],
1488 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1489 (memopv2i64 addr:$src2)))], 0>,
1492 let Constraints = "$src1 = $dst" in {
1493 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1494 !strconcat(OpcodeStr, "ps"), f128mem,
1495 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1496 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1497 (memopv2i64 addr:$src2)))]>, TB;
1499 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1500 !strconcat(OpcodeStr, "pd"), f128mem,
1501 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1502 (bc_v2i64 (v2f64 VR128:$src2))))],
1503 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1504 (memopv2i64 addr:$src2)))]>, TB, OpSize;
1508 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1510 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1512 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1513 !strconcat(OpcodeStr, "ps"), f256mem,
1514 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
1515 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1516 (memopv4i64 addr:$src2)))], 0>, VEX_4V;
1518 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1519 !strconcat(OpcodeStr, "pd"), f256mem,
1520 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1521 (bc_v4i64 (v4f64 VR256:$src2))))],
1522 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1523 (memopv4i64 addr:$src2)))], 0>,
1527 // AVX 256-bit packed logical ops forms
1528 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1529 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1530 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1531 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
1533 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1534 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1535 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1536 let isCommutable = 0 in
1537 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
1539 //===----------------------------------------------------------------------===//
1540 // SSE 1 & 2 - Arithmetic Instructions
1541 //===----------------------------------------------------------------------===//
1543 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1546 /// In addition, we also have a special variant of the scalar form here to
1547 /// represent the associated intrinsic operation. This form is unlike the
1548 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1549 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1551 /// These three forms can each be reg+reg or reg+mem.
1554 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1556 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1558 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1559 OpNode, FR32, f32mem, Is2Addr>, XS;
1560 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1561 OpNode, FR64, f64mem, Is2Addr>, XD;
1564 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1566 let mayLoad = 0 in {
1567 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1568 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1569 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1570 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1574 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1576 let mayLoad = 0 in {
1577 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1578 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1579 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1580 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1584 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1586 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1587 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1588 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1589 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1592 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1594 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1595 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1596 SSEPackedSingle, Is2Addr>, TB;
1598 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1599 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1600 SSEPackedDouble, Is2Addr>, TB, OpSize;
1603 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1604 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1605 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1606 SSEPackedSingle, 0>, TB;
1608 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1609 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1610 SSEPackedDouble, 0>, TB, OpSize;
1613 // Binary Arithmetic instructions
1614 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1615 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1616 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1617 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1618 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1619 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1620 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1621 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1623 let isCommutable = 0 in {
1624 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1625 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1626 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1627 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1628 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1629 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1630 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1631 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1632 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1633 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1634 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1635 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1636 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1637 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1638 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1639 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1640 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1641 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1642 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1643 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1646 let Constraints = "$src1 = $dst" in {
1647 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1648 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1649 basic_sse12_fp_binop_s_int<0x58, "add">;
1650 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1651 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1652 basic_sse12_fp_binop_s_int<0x59, "mul">;
1654 let isCommutable = 0 in {
1655 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1656 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1657 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1658 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1659 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1660 basic_sse12_fp_binop_s_int<0x5E, "div">;
1661 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1662 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1663 basic_sse12_fp_binop_s_int<0x5F, "max">,
1664 basic_sse12_fp_binop_p_int<0x5F, "max">;
1665 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1666 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1667 basic_sse12_fp_binop_s_int<0x5D, "min">,
1668 basic_sse12_fp_binop_p_int<0x5D, "min">;
1673 /// In addition, we also have a special variant of the scalar form here to
1674 /// represent the associated intrinsic operation. This form is unlike the
1675 /// plain scalar form, in that it takes an entire vector (instead of a
1676 /// scalar) and leaves the top elements undefined.
1678 /// And, we have a special variant form for a full-vector intrinsic form.
1680 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1681 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1682 SDNode OpNode, Intrinsic F32Int> {
1683 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1684 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1685 [(set FR32:$dst, (OpNode FR32:$src))]>;
1686 // For scalar unary operations, fold a load into the operation
1687 // only in OptForSize mode. It eliminates an instruction, but it also
1688 // eliminates a whole-register clobber (the load), so it introduces a
1689 // partial register update condition.
1690 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1691 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1692 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1693 Requires<[HasSSE1, OptForSize]>;
1694 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1695 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1696 [(set VR128:$dst, (F32Int VR128:$src))]>;
1697 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1698 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1699 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1702 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1703 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1704 SDNode OpNode, Intrinsic F32Int> {
1705 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1706 !strconcat(OpcodeStr,
1707 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1708 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1709 !strconcat(OpcodeStr,
1710 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1711 []>, XS, Requires<[HasAVX, OptForSize]>;
1712 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1713 !strconcat(OpcodeStr,
1714 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1715 [(set VR128:$dst, (F32Int VR128:$src))]>;
1716 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1717 !strconcat(OpcodeStr,
1718 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1719 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1722 /// sse1_fp_unop_p - SSE1 unops in packed form.
1723 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1724 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1725 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1726 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1727 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1728 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1729 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1732 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1733 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1734 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1735 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1736 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1737 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1738 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1739 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1742 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1743 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1744 Intrinsic V4F32Int> {
1745 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1746 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1747 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1748 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1749 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1750 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1753 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1754 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1755 Intrinsic V4F32Int> {
1756 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1757 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1758 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1759 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1760 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1761 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1764 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1765 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1766 SDNode OpNode, Intrinsic F64Int> {
1767 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1768 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1769 [(set FR64:$dst, (OpNode FR64:$src))]>;
1770 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1771 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1772 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1773 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1774 Requires<[HasSSE2, OptForSize]>;
1775 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1777 [(set VR128:$dst, (F64Int VR128:$src))]>;
1778 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1779 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1780 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1783 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1784 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1785 SDNode OpNode, Intrinsic F64Int> {
1786 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1787 !strconcat(OpcodeStr,
1788 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1789 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1790 (ins FR64:$src1, f64mem:$src2),
1791 !strconcat(OpcodeStr,
1792 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1793 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1794 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1795 [(set VR128:$dst, (F64Int VR128:$src))]>;
1796 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1797 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1798 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1801 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1802 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1804 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1805 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1806 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1807 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1808 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1809 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1812 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1813 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1814 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1815 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1816 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1817 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1818 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1819 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1822 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1823 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1824 Intrinsic V2F64Int> {
1825 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1826 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1827 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1828 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1829 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1830 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1833 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1834 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1835 Intrinsic V2F64Int> {
1836 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1837 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1838 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1839 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1840 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1841 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1844 let Predicates = [HasAVX] in {
1846 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1847 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1850 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1851 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1852 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1853 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1854 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1855 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1856 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1857 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1860 // Reciprocal approximations. Note that these typically require refinement
1861 // in order to obtain suitable precision.
1862 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1863 int_x86_sse_rsqrt_ss>, VEX_4V;
1864 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1865 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1866 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1867 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1869 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1871 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1872 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1873 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1874 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1878 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1879 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1880 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1881 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1882 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1883 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1885 // Reciprocal approximations. Note that these typically require refinement
1886 // in order to obtain suitable precision.
1887 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1888 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1889 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1890 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1891 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1892 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1894 // There is no f64 version of the reciprocal approximation instructions.
1896 //===----------------------------------------------------------------------===//
1897 // SSE 1 & 2 - Non-temporal stores
1898 //===----------------------------------------------------------------------===//
1900 let AddedComplexity = 400 in { // Prefer non-temporal versions
1901 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1902 (ins f128mem:$dst, VR128:$src),
1903 "movntps\t{$src, $dst|$dst, $src}",
1904 [(alignednontemporalstore (v4f32 VR128:$src),
1906 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1907 (ins f128mem:$dst, VR128:$src),
1908 "movntpd\t{$src, $dst|$dst, $src}",
1909 [(alignednontemporalstore (v2f64 VR128:$src),
1911 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1912 (ins f128mem:$dst, VR128:$src),
1913 "movntdq\t{$src, $dst|$dst, $src}",
1914 [(alignednontemporalstore (v2f64 VR128:$src),
1917 let ExeDomain = SSEPackedInt in
1918 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1919 (ins f128mem:$dst, VR128:$src),
1920 "movntdq\t{$src, $dst|$dst, $src}",
1921 [(alignednontemporalstore (v4f32 VR128:$src),
1924 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1925 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
1927 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1928 (ins f256mem:$dst, VR256:$src),
1929 "movntps\t{$src, $dst|$dst, $src}",
1930 [(alignednontemporalstore (v8f32 VR256:$src),
1932 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1933 (ins f256mem:$dst, VR256:$src),
1934 "movntpd\t{$src, $dst|$dst, $src}",
1935 [(alignednontemporalstore (v4f64 VR256:$src),
1937 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1938 (ins f256mem:$dst, VR256:$src),
1939 "movntdq\t{$src, $dst|$dst, $src}",
1940 [(alignednontemporalstore (v4f64 VR256:$src),
1942 let ExeDomain = SSEPackedInt in
1943 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1944 (ins f256mem:$dst, VR256:$src),
1945 "movntdq\t{$src, $dst|$dst, $src}",
1946 [(alignednontemporalstore (v8f32 VR256:$src),
1950 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1951 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1952 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
1953 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
1954 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
1955 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
1957 let AddedComplexity = 400 in { // Prefer non-temporal versions
1958 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1959 "movntps\t{$src, $dst|$dst, $src}",
1960 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1961 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1962 "movntpd\t{$src, $dst|$dst, $src}",
1963 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1965 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1966 "movntdq\t{$src, $dst|$dst, $src}",
1967 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1969 let ExeDomain = SSEPackedInt in
1970 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1971 "movntdq\t{$src, $dst|$dst, $src}",
1972 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1974 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1975 (MOVNTDQmr addr:$dst, VR128:$src)>;
1977 // There is no AVX form for instructions below this point
1978 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
1979 "movnti{l}\t{$src, $dst|$dst, $src}",
1980 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
1981 TB, Requires<[HasSSE2]>;
1982 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
1983 "movnti{q}\t{$src, $dst|$dst, $src}",
1984 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
1985 TB, Requires<[HasSSE2]>;
1988 //===----------------------------------------------------------------------===//
1989 // SSE 1 & 2 - Misc Instructions (No AVX form)
1990 //===----------------------------------------------------------------------===//
1992 // Prefetch intrinsic.
1993 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
1994 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
1995 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
1996 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
1997 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
1998 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
1999 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2000 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2002 // Load, store, and memory fence
2003 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2004 TB, Requires<[HasSSE1]>;
2005 def : Pat<(X86SFence), (SFENCE)>;
2007 // Alias instructions that map zero vector to pxor / xorp* for sse.
2008 // We set canFoldAsLoad because this can be converted to a constant-pool
2009 // load of an all-zeros value if folding it would be beneficial.
2010 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2011 // JIT implementation, it does not expand the instructions below like
2012 // X86MCInstLower does.
2013 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2014 isCodeGenOnly = 1 in {
2015 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2016 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2017 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2018 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2019 let ExeDomain = SSEPackedInt in
2020 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2021 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2024 // The same as done above but for AVX. The 128-bit versions are the
2025 // same, but re-encoded. The 256-bit does not support PI version.
2026 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2027 // JIT implementatioan, it does not expand the instructions below like
2028 // X86MCInstLower does.
2029 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2030 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2031 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2032 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2033 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2034 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2035 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2036 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2037 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2038 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2039 let ExeDomain = SSEPackedInt in
2040 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2041 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2044 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2045 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2046 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2048 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2049 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2051 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
2052 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
2053 // represent this instead of always zeroing SRC1. One possible solution is
2054 // to represent the instruction w/ something similar as the "$src1 = $dst"
2055 // constraint but without the tied operands.
2056 def : Pat<(extloadf32 addr:$src),
2057 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)), addr:$src)>,
2058 Requires<[HasAVX, OptForSpeed]>;
2060 //===----------------------------------------------------------------------===//
2061 // SSE 1 & 2 - Load/Store XCSR register
2062 //===----------------------------------------------------------------------===//
2064 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2065 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2066 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2067 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2069 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2070 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2071 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2072 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2074 //===---------------------------------------------------------------------===//
2075 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2076 //===---------------------------------------------------------------------===//
2078 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2080 let neverHasSideEffects = 1 in {
2081 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2082 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2083 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2084 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2086 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2087 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2088 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2089 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2091 let canFoldAsLoad = 1, mayLoad = 1 in {
2092 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2093 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2094 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2095 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2096 let Predicates = [HasAVX] in {
2097 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2098 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2099 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2100 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2104 let mayStore = 1 in {
2105 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2106 (ins i128mem:$dst, VR128:$src),
2107 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2108 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2109 (ins i256mem:$dst, VR256:$src),
2110 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2111 let Predicates = [HasAVX] in {
2112 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2113 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2114 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2115 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2119 let neverHasSideEffects = 1 in
2120 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2121 "movdqa\t{$src, $dst|$dst, $src}", []>;
2123 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2124 "movdqu\t{$src, $dst|$dst, $src}",
2125 []>, XS, Requires<[HasSSE2]>;
2127 let canFoldAsLoad = 1, mayLoad = 1 in {
2128 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2129 "movdqa\t{$src, $dst|$dst, $src}",
2130 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2131 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2132 "movdqu\t{$src, $dst|$dst, $src}",
2133 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2134 XS, Requires<[HasSSE2]>;
2137 let mayStore = 1 in {
2138 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2139 "movdqa\t{$src, $dst|$dst, $src}",
2140 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2141 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2142 "movdqu\t{$src, $dst|$dst, $src}",
2143 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2144 XS, Requires<[HasSSE2]>;
2147 // Intrinsic forms of MOVDQU load and store
2148 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2149 "vmovdqu\t{$src, $dst|$dst, $src}",
2150 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2151 XS, VEX, Requires<[HasAVX]>;
2153 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2154 "movdqu\t{$src, $dst|$dst, $src}",
2155 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2156 XS, Requires<[HasSSE2]>;
2158 } // ExeDomain = SSEPackedInt
2160 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2161 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2162 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2164 //===---------------------------------------------------------------------===//
2165 // SSE2 - Packed Integer Arithmetic Instructions
2166 //===---------------------------------------------------------------------===//
2168 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2170 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2171 bit IsCommutable = 0, bit Is2Addr = 1> {
2172 let isCommutable = IsCommutable in
2173 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2174 (ins VR128:$src1, VR128:$src2),
2176 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2177 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2178 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2179 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2180 (ins VR128:$src1, i128mem:$src2),
2182 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2183 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2184 [(set VR128:$dst, (IntId VR128:$src1,
2185 (bitconvert (memopv2i64 addr:$src2))))]>;
2188 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2189 string OpcodeStr, Intrinsic IntId,
2190 Intrinsic IntId2, bit Is2Addr = 1> {
2191 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2192 (ins VR128:$src1, VR128:$src2),
2194 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2195 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2196 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2197 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2198 (ins VR128:$src1, i128mem:$src2),
2200 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2201 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2202 [(set VR128:$dst, (IntId VR128:$src1,
2203 (bitconvert (memopv2i64 addr:$src2))))]>;
2204 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2205 (ins VR128:$src1, i32i8imm:$src2),
2207 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2208 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2209 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2212 /// PDI_binop_rm - Simple SSE2 binary operator.
2213 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2214 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2215 let isCommutable = IsCommutable in
2216 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2217 (ins VR128:$src1, VR128:$src2),
2219 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2220 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2221 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2222 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2223 (ins VR128:$src1, i128mem:$src2),
2225 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2226 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2227 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2228 (bitconvert (memopv2i64 addr:$src2)))))]>;
2231 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2233 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2234 /// to collapse (bitconvert VT to VT) into its operand.
2236 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2237 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, (v2i64 (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, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2253 } // ExeDomain = SSEPackedInt
2255 // 128-bit Integer Arithmetic
2257 let Predicates = [HasAVX] in {
2258 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2259 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2260 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2261 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2262 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2263 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2264 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2265 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2266 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2269 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2271 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2273 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2275 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2277 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2279 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2281 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2283 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2285 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2287 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2289 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2291 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2293 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2295 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2297 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2299 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2301 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2303 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2305 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2309 let Constraints = "$src1 = $dst" in {
2310 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2311 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2312 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2313 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2314 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2315 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2316 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2317 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2318 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2321 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2322 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2323 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2324 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2325 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2326 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2327 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2328 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2329 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2330 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2331 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2332 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2333 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2334 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2335 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2336 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2337 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2338 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2339 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2341 } // Constraints = "$src1 = $dst"
2343 //===---------------------------------------------------------------------===//
2344 // SSE2 - Packed Integer Logical Instructions
2345 //===---------------------------------------------------------------------===//
2347 let Predicates = [HasAVX] in {
2348 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2349 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2351 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2352 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2354 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2355 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2358 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2359 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2361 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2362 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2364 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2365 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2368 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2369 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2371 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2372 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2375 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2376 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2377 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2379 let ExeDomain = SSEPackedInt in {
2380 let neverHasSideEffects = 1 in {
2381 // 128-bit logical shifts.
2382 def VPSLLDQri : PDIi8<0x73, MRM7r,
2383 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2384 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2386 def VPSRLDQri : PDIi8<0x73, MRM3r,
2387 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2388 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2390 // PSRADQri doesn't exist in SSE[1-3].
2392 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2393 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2394 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2395 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2396 VR128:$src2)))]>, VEX_4V;
2398 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2399 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2400 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2401 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2402 (memopv2i64 addr:$src2))))]>,
2407 let Constraints = "$src1 = $dst" in {
2408 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2409 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2410 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2411 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2412 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2413 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2415 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2416 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2417 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2418 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2419 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2420 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2422 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2423 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2424 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2425 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2427 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2428 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2429 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2431 let ExeDomain = SSEPackedInt in {
2432 let neverHasSideEffects = 1 in {
2433 // 128-bit logical shifts.
2434 def PSLLDQri : PDIi8<0x73, MRM7r,
2435 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2436 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2437 def PSRLDQri : PDIi8<0x73, MRM3r,
2438 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2439 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2440 // PSRADQri doesn't exist in SSE[1-3].
2442 def PANDNrr : PDI<0xDF, MRMSrcReg,
2443 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2444 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2446 def PANDNrm : PDI<0xDF, MRMSrcMem,
2447 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2448 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2450 } // Constraints = "$src1 = $dst"
2452 let Predicates = [HasAVX] in {
2453 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2454 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2455 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2456 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2457 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2458 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2459 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2460 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2461 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2462 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2464 // Shift up / down and insert zero's.
2465 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2466 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2467 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2468 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2471 let Predicates = [HasSSE2] in {
2472 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2473 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2474 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2475 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2476 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2477 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2478 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2479 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2480 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2481 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2483 // Shift up / down and insert zero's.
2484 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2485 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2486 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2487 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2490 //===---------------------------------------------------------------------===//
2491 // SSE2 - Packed Integer Comparison Instructions
2492 //===---------------------------------------------------------------------===//
2494 let Predicates = [HasAVX] in {
2495 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2497 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2499 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2501 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2503 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2505 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2509 let Constraints = "$src1 = $dst" in {
2510 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2511 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2512 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2513 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2514 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2515 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2516 } // Constraints = "$src1 = $dst"
2518 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2519 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2520 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2521 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2522 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2523 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2524 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2525 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2526 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2527 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2528 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2529 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2531 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2532 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2533 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2534 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2535 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2536 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2537 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2538 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2539 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2540 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2541 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2542 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2544 //===---------------------------------------------------------------------===//
2545 // SSE2 - Packed Integer Pack Instructions
2546 //===---------------------------------------------------------------------===//
2548 let Predicates = [HasAVX] in {
2549 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2551 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2553 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2557 let Constraints = "$src1 = $dst" in {
2558 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2559 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2560 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2561 } // Constraints = "$src1 = $dst"
2563 //===---------------------------------------------------------------------===//
2564 // SSE2 - Packed Integer Shuffle Instructions
2565 //===---------------------------------------------------------------------===//
2567 let ExeDomain = SSEPackedInt in {
2568 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2570 def ri : Ii8<0x70, MRMSrcReg,
2571 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2572 !strconcat(OpcodeStr,
2573 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2574 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2576 def mi : Ii8<0x70, MRMSrcMem,
2577 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2578 !strconcat(OpcodeStr,
2579 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2580 [(set VR128:$dst, (vt (pshuf_frag:$src2
2581 (bc_frag (memopv2i64 addr:$src1)),
2584 } // ExeDomain = SSEPackedInt
2586 let Predicates = [HasAVX] in {
2587 let AddedComplexity = 5 in
2588 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2591 // SSE2 with ImmT == Imm8 and XS prefix.
2592 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2595 // SSE2 with ImmT == Imm8 and XD prefix.
2596 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2600 let Predicates = [HasSSE2] in {
2601 let AddedComplexity = 5 in
2602 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2604 // SSE2 with ImmT == Imm8 and XS prefix.
2605 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2607 // SSE2 with ImmT == Imm8 and XD prefix.
2608 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2611 //===---------------------------------------------------------------------===//
2612 // SSE2 - Packed Integer Unpack Instructions
2613 //===---------------------------------------------------------------------===//
2615 let ExeDomain = SSEPackedInt in {
2616 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2617 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2618 def rr : PDI<opc, MRMSrcReg,
2619 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2621 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2622 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2623 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2624 def rm : PDI<opc, MRMSrcMem,
2625 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2627 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2628 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2629 [(set VR128:$dst, (unp_frag VR128:$src1,
2630 (bc_frag (memopv2i64
2634 let Predicates = [HasAVX] in {
2635 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2637 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2639 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2642 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2643 /// knew to collapse (bitconvert VT to VT) into its operand.
2644 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2645 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2646 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2648 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2649 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2650 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2651 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2653 (v2i64 (unpckl VR128:$src1,
2654 (memopv2i64 addr:$src2))))]>, VEX_4V;
2656 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2658 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2660 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2663 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2664 /// knew to collapse (bitconvert VT to VT) into its operand.
2665 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2666 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2667 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2669 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2670 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2671 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2672 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2674 (v2i64 (unpckh VR128:$src1,
2675 (memopv2i64 addr:$src2))))]>, VEX_4V;
2678 let Constraints = "$src1 = $dst" in {
2679 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2680 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2681 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2683 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2684 /// knew to collapse (bitconvert VT to VT) into its operand.
2685 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2686 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2687 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2689 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2690 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2691 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2692 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2694 (v2i64 (unpckl VR128:$src1,
2695 (memopv2i64 addr:$src2))))]>;
2697 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2698 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2699 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2701 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2702 /// knew to collapse (bitconvert VT to VT) into its operand.
2703 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2704 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2705 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2707 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2708 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2709 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2710 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2712 (v2i64 (unpckh VR128:$src1,
2713 (memopv2i64 addr:$src2))))]>;
2716 } // ExeDomain = SSEPackedInt
2718 //===---------------------------------------------------------------------===//
2719 // SSE2 - Packed Integer Extract and Insert
2720 //===---------------------------------------------------------------------===//
2722 let ExeDomain = SSEPackedInt in {
2723 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2724 def rri : Ii8<0xC4, MRMSrcReg,
2725 (outs VR128:$dst), (ins VR128:$src1,
2726 GR32:$src2, i32i8imm:$src3),
2728 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2729 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2731 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2732 def rmi : Ii8<0xC4, MRMSrcMem,
2733 (outs VR128:$dst), (ins VR128:$src1,
2734 i16mem:$src2, i32i8imm:$src3),
2736 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2737 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2739 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2744 let Predicates = [HasAVX] in
2745 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2746 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2747 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2748 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2749 imm:$src2))]>, OpSize, VEX;
2750 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2751 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2752 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2753 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2757 let Predicates = [HasAVX] in {
2758 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2759 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2760 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2761 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2762 []>, OpSize, VEX_4V;
2765 let Constraints = "$src1 = $dst" in
2766 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2768 } // ExeDomain = SSEPackedInt
2770 //===---------------------------------------------------------------------===//
2771 // SSE2 - Packed Mask Creation
2772 //===---------------------------------------------------------------------===//
2774 let ExeDomain = SSEPackedInt in {
2776 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2777 "pmovmskb\t{$src, $dst|$dst, $src}",
2778 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2779 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2780 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2781 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2782 "pmovmskb\t{$src, $dst|$dst, $src}",
2783 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2785 } // ExeDomain = SSEPackedInt
2787 //===---------------------------------------------------------------------===//
2788 // SSE2 - Conditional Store
2789 //===---------------------------------------------------------------------===//
2791 let ExeDomain = SSEPackedInt in {
2794 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2795 (ins VR128:$src, VR128:$mask),
2796 "maskmovdqu\t{$mask, $src|$src, $mask}",
2797 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2799 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2800 (ins VR128:$src, VR128:$mask),
2801 "maskmovdqu\t{$mask, $src|$src, $mask}",
2802 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2805 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2806 "maskmovdqu\t{$mask, $src|$src, $mask}",
2807 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2809 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2810 "maskmovdqu\t{$mask, $src|$src, $mask}",
2811 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2813 } // ExeDomain = SSEPackedInt
2815 //===---------------------------------------------------------------------===//
2816 // SSE2 - Move Doubleword
2817 //===---------------------------------------------------------------------===//
2819 // Move Int Doubleword to Packed Double Int
2820 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2821 "movd\t{$src, $dst|$dst, $src}",
2823 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2824 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2825 "movd\t{$src, $dst|$dst, $src}",
2827 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2829 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2830 "movd\t{$src, $dst|$dst, $src}",
2832 (v4i32 (scalar_to_vector GR32:$src)))]>;
2833 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2834 "movd\t{$src, $dst|$dst, $src}",
2836 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2837 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2838 "mov{d|q}\t{$src, $dst|$dst, $src}",
2840 (v2i64 (scalar_to_vector GR64:$src)))]>;
2841 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2842 "mov{d|q}\t{$src, $dst|$dst, $src}",
2843 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2846 // Move Int Doubleword to Single Scalar
2847 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2848 "movd\t{$src, $dst|$dst, $src}",
2849 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2851 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2852 "movd\t{$src, $dst|$dst, $src}",
2853 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2855 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2856 "movd\t{$src, $dst|$dst, $src}",
2857 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2859 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2860 "movd\t{$src, $dst|$dst, $src}",
2861 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2863 // Move Packed Doubleword Int to Packed Double Int
2864 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2865 "movd\t{$src, $dst|$dst, $src}",
2866 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2868 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2869 (ins i32mem:$dst, VR128:$src),
2870 "movd\t{$src, $dst|$dst, $src}",
2871 [(store (i32 (vector_extract (v4i32 VR128:$src),
2872 (iPTR 0))), addr:$dst)]>, VEX;
2873 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2874 "movd\t{$src, $dst|$dst, $src}",
2875 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2877 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2878 "movd\t{$src, $dst|$dst, $src}",
2879 [(store (i32 (vector_extract (v4i32 VR128:$src),
2880 (iPTR 0))), addr:$dst)]>;
2882 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2883 "mov{d|q}\t{$src, $dst|$dst, $src}",
2884 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2886 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2887 "movq\t{$src, $dst|$dst, $src}",
2888 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2890 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2891 "mov{d|q}\t{$src, $dst|$dst, $src}",
2892 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2893 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2894 "movq\t{$src, $dst|$dst, $src}",
2895 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2897 // Move Scalar Single to Double Int
2898 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2899 "movd\t{$src, $dst|$dst, $src}",
2900 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2901 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2902 "movd\t{$src, $dst|$dst, $src}",
2903 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2904 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2905 "movd\t{$src, $dst|$dst, $src}",
2906 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2907 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2908 "movd\t{$src, $dst|$dst, $src}",
2909 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2911 // movd / movq to XMM register zero-extends
2912 let AddedComplexity = 15 in {
2913 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2914 "movd\t{$src, $dst|$dst, $src}",
2915 [(set VR128:$dst, (v4i32 (X86vzmovl
2916 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2918 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2919 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2920 [(set VR128:$dst, (v2i64 (X86vzmovl
2921 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2924 let AddedComplexity = 15 in {
2925 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2926 "movd\t{$src, $dst|$dst, $src}",
2927 [(set VR128:$dst, (v4i32 (X86vzmovl
2928 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2929 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2930 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2931 [(set VR128:$dst, (v2i64 (X86vzmovl
2932 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2935 let AddedComplexity = 20 in {
2936 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2937 "movd\t{$src, $dst|$dst, $src}",
2939 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2940 (loadi32 addr:$src))))))]>,
2942 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2943 "movd\t{$src, $dst|$dst, $src}",
2945 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2946 (loadi32 addr:$src))))))]>;
2948 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2949 (MOVZDI2PDIrm addr:$src)>;
2950 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2951 (MOVZDI2PDIrm addr:$src)>;
2952 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2953 (MOVZDI2PDIrm addr:$src)>;
2956 // These are the correct encodings of the instructions so that we know how to
2957 // read correct assembly, even though we continue to emit the wrong ones for
2958 // compatibility with Darwin's buggy assembler.
2959 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2960 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
2961 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2962 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
2963 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2964 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
2965 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2966 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
2967 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2968 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2969 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2970 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2972 //===---------------------------------------------------------------------===//
2973 // SSE2 - Move Quadword
2974 //===---------------------------------------------------------------------===//
2976 // Move Quadword Int to Packed Quadword Int
2977 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2978 "vmovq\t{$src, $dst|$dst, $src}",
2980 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2981 VEX, Requires<[HasAVX]>;
2982 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
2983 "movq\t{$src, $dst|$dst, $src}",
2985 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
2986 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
2988 // Move Packed Quadword Int to Quadword Int
2989 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2990 "movq\t{$src, $dst|$dst, $src}",
2991 [(store (i64 (vector_extract (v2i64 VR128:$src),
2992 (iPTR 0))), addr:$dst)]>, VEX;
2993 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
2994 "movq\t{$src, $dst|$dst, $src}",
2995 [(store (i64 (vector_extract (v2i64 VR128:$src),
2996 (iPTR 0))), addr:$dst)]>;
2998 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
2999 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3001 // Store / copy lower 64-bits of a XMM register.
3002 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3003 "movq\t{$src, $dst|$dst, $src}",
3004 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3005 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3006 "movq\t{$src, $dst|$dst, $src}",
3007 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3009 let AddedComplexity = 20 in
3010 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3011 "vmovq\t{$src, $dst|$dst, $src}",
3013 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3014 (loadi64 addr:$src))))))]>,
3015 XS, VEX, Requires<[HasAVX]>;
3017 let AddedComplexity = 20 in {
3018 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3019 "movq\t{$src, $dst|$dst, $src}",
3021 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3022 (loadi64 addr:$src))))))]>,
3023 XS, Requires<[HasSSE2]>;
3025 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3026 (MOVZQI2PQIrm addr:$src)>;
3027 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3028 (MOVZQI2PQIrm addr:$src)>;
3029 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3032 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3033 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3034 let AddedComplexity = 15 in
3035 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3036 "vmovq\t{$src, $dst|$dst, $src}",
3037 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3038 XS, VEX, Requires<[HasAVX]>;
3039 let AddedComplexity = 15 in
3040 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3041 "movq\t{$src, $dst|$dst, $src}",
3042 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3043 XS, Requires<[HasSSE2]>;
3045 let AddedComplexity = 20 in
3046 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3047 "vmovq\t{$src, $dst|$dst, $src}",
3048 [(set VR128:$dst, (v2i64 (X86vzmovl
3049 (loadv2i64 addr:$src))))]>,
3050 XS, VEX, Requires<[HasAVX]>;
3051 let AddedComplexity = 20 in {
3052 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3053 "movq\t{$src, $dst|$dst, $src}",
3054 [(set VR128:$dst, (v2i64 (X86vzmovl
3055 (loadv2i64 addr:$src))))]>,
3056 XS, Requires<[HasSSE2]>;
3058 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3059 (MOVZPQILo2PQIrm addr:$src)>;
3062 // Instructions to match in the assembler
3063 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3064 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3065 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3066 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3067 // Recognize "movd" with GR64 destination, but encode as a "movq"
3068 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3069 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3071 // Instructions for the disassembler
3072 // xr = XMM register
3075 let Predicates = [HasAVX] in
3076 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3077 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3078 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3079 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3081 //===---------------------------------------------------------------------===//
3082 // SSE2 - Misc Instructions
3083 //===---------------------------------------------------------------------===//
3086 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3087 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3088 TB, Requires<[HasSSE2]>;
3090 // Load, store, and memory fence
3091 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3092 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3093 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3094 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3095 def : Pat<(X86LFence), (LFENCE)>;
3096 def : Pat<(X86MFence), (MFENCE)>;
3099 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3100 // was introduced with SSE2, it's backward compatible.
3101 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3103 // Alias instructions that map zero vector to pxor / xorp* for sse.
3104 // We set canFoldAsLoad because this can be converted to a constant-pool
3105 // load of an all-ones value if folding it would be beneficial.
3106 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3107 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3108 // FIXME: Change encoding to pseudo.
3109 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3110 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3112 //===---------------------------------------------------------------------===//
3113 // SSE3 - Conversion Instructions
3114 //===---------------------------------------------------------------------===//
3116 // Convert Packed Double FP to Packed DW Integers
3117 let Predicates = [HasAVX] in {
3118 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3119 // register, but the same isn't true when using memory operands instead.
3120 // Provide other assembly rr and rm forms to address this explicitly.
3121 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3122 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3123 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3124 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3127 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3128 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3129 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3130 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3133 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3134 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3135 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3136 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3139 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3140 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3141 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3142 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3144 // Convert Packed DW Integers to Packed Double FP
3145 let Predicates = [HasAVX] in {
3146 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3147 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3148 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3149 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3150 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3151 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3152 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3153 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3156 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3157 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3158 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3159 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3161 // AVX 256-bit register conversion intrinsics
3162 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3163 (VCVTDQ2PDYrr VR128:$src)>;
3164 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3165 (VCVTDQ2PDYrm addr:$src)>;
3167 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3168 (VCVTPD2DQYrr VR256:$src)>;
3169 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3170 (VCVTPD2DQYrm addr:$src)>;
3172 //===---------------------------------------------------------------------===//
3173 // SSE3 - Move Instructions
3174 //===---------------------------------------------------------------------===//
3176 // Replicate Single FP
3177 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3178 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3179 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3180 [(set VR128:$dst, (v4f32 (rep_frag
3181 VR128:$src, (undef))))]>;
3182 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3183 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3184 [(set VR128:$dst, (rep_frag
3185 (memopv4f32 addr:$src), (undef)))]>;
3188 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3190 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3191 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3192 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3193 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3196 let Predicates = [HasAVX] in {
3197 // FIXME: Merge above classes when we have patterns for the ymm version
3198 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3199 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3200 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3201 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3203 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3204 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3206 // Replicate Double FP
3207 multiclass sse3_replicate_dfp<string OpcodeStr> {
3208 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3209 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3210 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3211 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3212 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3214 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3218 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3219 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3220 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3222 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3223 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3227 let Predicates = [HasAVX] in {
3228 // FIXME: Merge above classes when we have patterns for the ymm version
3229 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3230 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3232 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3234 // Move Unaligned Integer
3235 let Predicates = [HasAVX] in {
3236 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3237 "vlddqu\t{$src, $dst|$dst, $src}",
3238 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3239 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3240 "vlddqu\t{$src, $dst|$dst, $src}",
3241 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3243 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3244 "lddqu\t{$src, $dst|$dst, $src}",
3245 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3247 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3249 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3251 // Several Move patterns
3252 let AddedComplexity = 5 in {
3253 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3254 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3255 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3256 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3257 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3258 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3259 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3260 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3263 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3264 let AddedComplexity = 15 in
3265 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3266 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3267 let AddedComplexity = 20 in
3268 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3269 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3271 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3272 let AddedComplexity = 15 in
3273 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3274 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3275 let AddedComplexity = 20 in
3276 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3277 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3279 //===---------------------------------------------------------------------===//
3280 // SSE3 - Arithmetic
3281 //===---------------------------------------------------------------------===//
3283 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3284 X86MemOperand x86memop, bit Is2Addr = 1> {
3285 def rr : I<0xD0, MRMSrcReg,
3286 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3288 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3289 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3290 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3291 def rm : I<0xD0, MRMSrcMem,
3292 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3294 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3295 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3296 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3299 let Predicates = [HasAVX],
3300 ExeDomain = SSEPackedDouble in {
3301 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3302 f128mem, 0>, TB, XD, VEX_4V;
3303 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3304 f128mem, 0>, TB, OpSize, VEX_4V;
3305 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3306 f256mem, 0>, TB, XD, VEX_4V;
3307 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3308 f256mem, 0>, TB, OpSize, VEX_4V;
3310 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3311 ExeDomain = SSEPackedDouble in {
3312 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3314 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3315 f128mem>, TB, OpSize;
3318 //===---------------------------------------------------------------------===//
3319 // SSE3 Instructions
3320 //===---------------------------------------------------------------------===//
3323 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3324 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3325 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3327 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3328 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3329 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3331 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3333 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3334 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3335 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3337 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3338 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3339 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3343 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3345 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3347 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3349 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3352 let Predicates = [HasAVX] in {
3353 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3354 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3355 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3356 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3357 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3358 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3359 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3360 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3361 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3362 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3363 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3364 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3365 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3366 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3367 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3368 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3371 let Constraints = "$src1 = $dst" in {
3372 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3373 int_x86_sse3_hadd_ps>;
3374 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3375 int_x86_sse3_hadd_pd>;
3376 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3377 int_x86_sse3_hsub_ps>;
3378 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3379 int_x86_sse3_hsub_pd>;
3382 //===---------------------------------------------------------------------===//
3383 // SSSE3 - Packed Absolute Instructions
3384 //===---------------------------------------------------------------------===//
3387 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3388 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3389 PatFrag mem_frag128, Intrinsic IntId128> {
3390 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3392 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3393 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3396 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3398 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3401 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3404 let Predicates = [HasAVX] in {
3405 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3406 int_x86_ssse3_pabs_b_128>, VEX;
3407 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3408 int_x86_ssse3_pabs_w_128>, VEX;
3409 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3410 int_x86_ssse3_pabs_d_128>, VEX;
3413 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3414 int_x86_ssse3_pabs_b_128>;
3415 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3416 int_x86_ssse3_pabs_w_128>;
3417 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3418 int_x86_ssse3_pabs_d_128>;
3420 //===---------------------------------------------------------------------===//
3421 // SSSE3 - Packed Binary Operator Instructions
3422 //===---------------------------------------------------------------------===//
3424 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3425 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3426 PatFrag mem_frag128, Intrinsic IntId128,
3428 let isCommutable = 1 in
3429 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3430 (ins VR128:$src1, VR128:$src2),
3432 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3433 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3434 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3436 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3437 (ins VR128:$src1, i128mem:$src2),
3439 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3440 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3442 (IntId128 VR128:$src1,
3443 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3446 let Predicates = [HasAVX] in {
3447 let isCommutable = 0 in {
3448 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3449 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3450 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3451 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3452 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3453 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3454 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3455 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3456 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3457 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3458 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3459 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3460 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3461 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3462 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3463 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3464 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3465 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3466 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3467 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3468 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3469 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3471 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3472 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3475 // None of these have i8 immediate fields.
3476 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3477 let isCommutable = 0 in {
3478 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3479 int_x86_ssse3_phadd_w_128>;
3480 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3481 int_x86_ssse3_phadd_d_128>;
3482 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3483 int_x86_ssse3_phadd_sw_128>;
3484 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3485 int_x86_ssse3_phsub_w_128>;
3486 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3487 int_x86_ssse3_phsub_d_128>;
3488 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3489 int_x86_ssse3_phsub_sw_128>;
3490 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3491 int_x86_ssse3_pmadd_ub_sw_128>;
3492 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3493 int_x86_ssse3_pshuf_b_128>;
3494 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3495 int_x86_ssse3_psign_b_128>;
3496 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3497 int_x86_ssse3_psign_w_128>;
3498 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3499 int_x86_ssse3_psign_d_128>;
3501 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3502 int_x86_ssse3_pmul_hr_sw_128>;
3505 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3506 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3507 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3508 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3510 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3511 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3512 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3513 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3514 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3515 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3517 //===---------------------------------------------------------------------===//
3518 // SSSE3 - Packed Align Instruction Patterns
3519 //===---------------------------------------------------------------------===//
3521 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3522 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3523 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3525 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3527 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3529 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3530 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3532 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3534 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3538 let Predicates = [HasAVX] in
3539 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3540 let Constraints = "$src1 = $dst" in
3541 defm PALIGN : ssse3_palign<"palignr">;
3543 let AddedComplexity = 5 in {
3544 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3545 (PALIGNR128rr VR128:$src2, VR128:$src1,
3546 (SHUFFLE_get_palign_imm VR128:$src3))>,
3547 Requires<[HasSSSE3]>;
3548 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3549 (PALIGNR128rr VR128:$src2, VR128:$src1,
3550 (SHUFFLE_get_palign_imm VR128:$src3))>,
3551 Requires<[HasSSSE3]>;
3552 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3553 (PALIGNR128rr VR128:$src2, VR128:$src1,
3554 (SHUFFLE_get_palign_imm VR128:$src3))>,
3555 Requires<[HasSSSE3]>;
3556 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3557 (PALIGNR128rr VR128:$src2, VR128:$src1,
3558 (SHUFFLE_get_palign_imm VR128:$src3))>,
3559 Requires<[HasSSSE3]>;
3562 //===---------------------------------------------------------------------===//
3563 // SSSE3 Misc Instructions
3564 //===---------------------------------------------------------------------===//
3566 // Thread synchronization
3567 let usesCustomInserter = 1 in {
3568 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3569 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3570 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3571 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3574 let Uses = [EAX, ECX, EDX] in
3575 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3576 Requires<[HasSSE3]>;
3577 let Uses = [ECX, EAX] in
3578 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3579 Requires<[HasSSE3]>;
3581 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3582 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3584 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3585 Requires<[In32BitMode]>;
3586 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3587 Requires<[In64BitMode]>;
3589 //===---------------------------------------------------------------------===//
3590 // Non-Instruction Patterns
3591 //===---------------------------------------------------------------------===//
3593 // extload f32 -> f64. This matches load+fextend because we have a hack in
3594 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3596 // Since these loads aren't folded into the fextend, we have to match it
3598 let Predicates = [HasSSE2] in
3599 def : Pat<(fextend (loadf32 addr:$src)),
3600 (CVTSS2SDrm addr:$src)>;
3602 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
3603 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
3604 // represent this instead of always zeroing SRC1. One possible solution is
3605 // to represent the instruction w/ something similar as the "$src1 = $dst"
3606 // constraint but without the tied operands.
3607 let Predicates = [HasAVX] in
3608 def : Pat<(fextend (loadf32 addr:$src)),
3609 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)),
3613 let Predicates = [HasXMMInt] in {
3614 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3615 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3616 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3617 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3618 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3619 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3620 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3621 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3622 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3623 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3624 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3625 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3626 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3627 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3628 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3629 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3630 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3631 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3632 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3633 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3634 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3635 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3636 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3637 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3638 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3639 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3640 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3641 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3642 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3643 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3646 let Predicates = [HasAVX] in {
3647 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3648 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
3649 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
3650 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
3651 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
3654 // Move scalar to XMM zero-extended
3655 // movd to XMM register zero-extends
3656 let AddedComplexity = 15 in {
3657 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3658 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3659 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3660 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3661 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3662 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3663 (MOVSSrr (v4f32 (V_SET0PS)),
3664 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3665 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3666 (MOVSSrr (v4i32 (V_SET0PI)),
3667 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3670 // Splat v2f64 / v2i64
3671 let AddedComplexity = 10 in {
3672 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3673 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3674 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3675 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3676 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3677 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3678 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3679 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3682 // Special unary SHUFPSrri case.
3683 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3684 (SHUFPSrri VR128:$src1, VR128:$src1,
3685 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3686 let AddedComplexity = 5 in
3687 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3688 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3689 Requires<[HasSSE2]>;
3690 // Special unary SHUFPDrri case.
3691 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3692 (SHUFPDrri VR128:$src1, VR128:$src1,
3693 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3694 Requires<[HasSSE2]>;
3695 // Special unary SHUFPDrri case.
3696 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3697 (SHUFPDrri VR128:$src1, VR128:$src1,
3698 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3699 Requires<[HasSSE2]>;
3700 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3701 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3702 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3703 Requires<[HasSSE2]>;
3705 // Special binary v4i32 shuffle cases with SHUFPS.
3706 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3707 (SHUFPSrri VR128:$src1, VR128:$src2,
3708 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3709 Requires<[HasSSE2]>;
3710 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3711 (SHUFPSrmi VR128:$src1, addr:$src2,
3712 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3713 Requires<[HasSSE2]>;
3714 // Special binary v2i64 shuffle cases using SHUFPDrri.
3715 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3716 (SHUFPDrri VR128:$src1, VR128:$src2,
3717 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3718 Requires<[HasSSE2]>;
3720 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3721 let AddedComplexity = 15 in {
3722 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3723 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3724 Requires<[OptForSpeed, HasSSE2]>;
3725 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3726 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3727 Requires<[OptForSpeed, HasSSE2]>;
3729 let AddedComplexity = 10 in {
3730 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3731 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3732 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3733 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3734 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3735 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3736 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3737 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3740 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3741 let AddedComplexity = 15 in {
3742 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3743 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3744 Requires<[OptForSpeed, HasSSE2]>;
3745 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3746 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3747 Requires<[OptForSpeed, HasSSE2]>;
3749 let AddedComplexity = 10 in {
3750 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3751 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3752 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3753 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3754 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3755 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3756 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3757 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3760 let AddedComplexity = 20 in {
3761 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3762 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3763 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3765 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3766 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3767 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3769 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3770 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3771 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3772 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3773 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3776 let AddedComplexity = 20 in {
3777 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3778 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3779 (MOVLPSrm VR128:$src1, addr:$src2)>;
3780 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3781 (MOVLPDrm VR128:$src1, addr:$src2)>;
3782 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3783 (MOVLPSrm VR128:$src1, addr:$src2)>;
3784 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3785 (MOVLPDrm VR128:$src1, addr:$src2)>;
3788 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3789 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3790 (MOVLPSmr addr:$src1, VR128:$src2)>;
3791 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3792 (MOVLPDmr addr:$src1, VR128:$src2)>;
3793 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3795 (MOVLPSmr addr:$src1, VR128:$src2)>;
3796 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3797 (MOVLPDmr addr:$src1, VR128:$src2)>;
3799 let AddedComplexity = 15 in {
3800 // Setting the lowest element in the vector.
3801 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3802 (MOVSSrr (v4i32 VR128:$src1),
3803 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3804 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3805 (MOVSDrr (v2i64 VR128:$src1),
3806 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3808 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3809 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3810 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3811 Requires<[HasSSE2]>;
3812 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3813 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3814 Requires<[HasSSE2]>;
3817 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3818 // fall back to this for SSE1)
3819 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3820 (SHUFPSrri VR128:$src2, VR128:$src1,
3821 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3823 // Set lowest element and zero upper elements.
3824 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3825 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3827 // vector -> vector casts
3828 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3829 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3830 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3831 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3833 // Use movaps / movups for SSE integer load / store (one byte shorter).
3834 let Predicates = [HasSSE1] in {
3835 def : Pat<(alignedloadv4i32 addr:$src),
3836 (MOVAPSrm addr:$src)>;
3837 def : Pat<(loadv4i32 addr:$src),
3838 (MOVUPSrm addr:$src)>;
3839 def : Pat<(alignedloadv2i64 addr:$src),
3840 (MOVAPSrm addr:$src)>;
3841 def : Pat<(loadv2i64 addr:$src),
3842 (MOVUPSrm addr:$src)>;
3844 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3845 (MOVAPSmr addr:$dst, VR128:$src)>;
3846 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3847 (MOVAPSmr addr:$dst, VR128:$src)>;
3848 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3849 (MOVAPSmr addr:$dst, VR128:$src)>;
3850 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3851 (MOVAPSmr addr:$dst, VR128:$src)>;
3852 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3853 (MOVUPSmr addr:$dst, VR128:$src)>;
3854 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3855 (MOVUPSmr addr:$dst, VR128:$src)>;
3856 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3857 (MOVUPSmr addr:$dst, VR128:$src)>;
3858 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3859 (MOVUPSmr addr:$dst, VR128:$src)>;
3862 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3863 let Predicates = [HasAVX] in {
3864 def : Pat<(alignedloadv4i32 addr:$src),
3865 (VMOVAPSrm addr:$src)>;
3866 def : Pat<(loadv4i32 addr:$src),
3867 (VMOVUPSrm addr:$src)>;
3868 def : Pat<(alignedloadv2i64 addr:$src),
3869 (VMOVAPSrm addr:$src)>;
3870 def : Pat<(loadv2i64 addr:$src),
3871 (VMOVUPSrm addr:$src)>;
3873 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3874 (VMOVAPSmr addr:$dst, VR128:$src)>;
3875 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3876 (VMOVAPSmr addr:$dst, VR128:$src)>;
3877 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3878 (VMOVAPSmr addr:$dst, VR128:$src)>;
3879 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3880 (VMOVAPSmr addr:$dst, VR128:$src)>;
3881 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3882 (VMOVUPSmr addr:$dst, VR128:$src)>;
3883 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3884 (VMOVUPSmr addr:$dst, VR128:$src)>;
3885 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3886 (VMOVUPSmr addr:$dst, VR128:$src)>;
3887 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3888 (VMOVUPSmr addr:$dst, VR128:$src)>;
3891 //===----------------------------------------------------------------------===//
3892 // SSE4.1 - Packed Move with Sign/Zero Extend
3893 //===----------------------------------------------------------------------===//
3895 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3896 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3897 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3898 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3900 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3901 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3903 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3907 let Predicates = [HasAVX] in {
3908 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3910 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3912 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3914 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3916 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3918 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3922 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3923 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3924 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3925 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3926 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3927 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3929 // Common patterns involving scalar load.
3930 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3931 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3932 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3933 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3935 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3936 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3937 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3938 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3940 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3941 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3942 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3943 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3945 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3946 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3947 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3948 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3950 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3951 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3952 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3953 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3955 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3956 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3957 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3958 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3961 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3962 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3963 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3964 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3966 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3967 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3969 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
3973 let Predicates = [HasAVX] in {
3974 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
3976 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
3978 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
3980 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
3984 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
3985 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
3986 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
3987 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
3989 // Common patterns involving scalar load
3990 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
3991 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
3992 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
3993 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
3995 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
3996 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
3997 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
3998 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4001 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4002 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4003 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4004 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4006 // Expecting a i16 load any extended to i32 value.
4007 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4008 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4009 [(set VR128:$dst, (IntId (bitconvert
4010 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4014 let Predicates = [HasAVX] in {
4015 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4017 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4020 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4021 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4023 // Common patterns involving scalar load
4024 def : Pat<(int_x86_sse41_pmovsxbq
4025 (bitconvert (v4i32 (X86vzmovl
4026 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4027 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4029 def : Pat<(int_x86_sse41_pmovzxbq
4030 (bitconvert (v4i32 (X86vzmovl
4031 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4032 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4034 //===----------------------------------------------------------------------===//
4035 // SSE4.1 - Extract Instructions
4036 //===----------------------------------------------------------------------===//
4038 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4039 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4040 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4041 (ins VR128:$src1, i32i8imm:$src2),
4042 !strconcat(OpcodeStr,
4043 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4044 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4046 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4047 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4048 !strconcat(OpcodeStr,
4049 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4052 // There's an AssertZext in the way of writing the store pattern
4053 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4056 let Predicates = [HasAVX] in {
4057 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4058 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4059 (ins VR128:$src1, i32i8imm:$src2),
4060 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4063 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4066 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4067 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4068 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4069 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4070 !strconcat(OpcodeStr,
4071 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4074 // There's an AssertZext in the way of writing the store pattern
4075 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4078 let Predicates = [HasAVX] in
4079 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4081 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4084 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4085 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4086 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4087 (ins VR128:$src1, i32i8imm:$src2),
4088 !strconcat(OpcodeStr,
4089 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4091 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4092 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4093 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4094 !strconcat(OpcodeStr,
4095 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4096 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4097 addr:$dst)]>, OpSize;
4100 let Predicates = [HasAVX] in
4101 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4103 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4105 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4106 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4107 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4108 (ins VR128:$src1, i32i8imm:$src2),
4109 !strconcat(OpcodeStr,
4110 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4112 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4113 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4114 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4115 !strconcat(OpcodeStr,
4116 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4117 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4118 addr:$dst)]>, OpSize, REX_W;
4121 let Predicates = [HasAVX] in
4122 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4124 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4126 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4128 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4129 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4130 (ins VR128:$src1, i32i8imm:$src2),
4131 !strconcat(OpcodeStr,
4132 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4134 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4136 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4137 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4138 !strconcat(OpcodeStr,
4139 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4140 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4141 addr:$dst)]>, OpSize;
4144 let Predicates = [HasAVX] in {
4145 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4146 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4147 (ins VR128:$src1, i32i8imm:$src2),
4148 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4151 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4153 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4154 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4157 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4158 Requires<[HasSSE41]>;
4160 //===----------------------------------------------------------------------===//
4161 // SSE4.1 - Insert Instructions
4162 //===----------------------------------------------------------------------===//
4164 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4165 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4166 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4168 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4170 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4172 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4173 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4174 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4176 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4178 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4180 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4181 imm:$src3))]>, OpSize;
4184 let Predicates = [HasAVX] in
4185 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4186 let Constraints = "$src1 = $dst" in
4187 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4189 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4190 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4191 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4193 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4195 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4197 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4199 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4200 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4202 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4204 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4206 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4207 imm:$src3)))]>, OpSize;
4210 let Predicates = [HasAVX] in
4211 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4212 let Constraints = "$src1 = $dst" in
4213 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4215 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4216 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4217 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4219 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4221 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4223 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4225 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4226 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4228 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4230 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4232 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4233 imm:$src3)))]>, OpSize;
4236 let Predicates = [HasAVX] in
4237 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4238 let Constraints = "$src1 = $dst" in
4239 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4241 // insertps has a few different modes, there's the first two here below which
4242 // are optimized inserts that won't zero arbitrary elements in the destination
4243 // vector. The next one matches the intrinsic and could zero arbitrary elements
4244 // in the target vector.
4245 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4246 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4247 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4249 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4251 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4253 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4255 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4256 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4258 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4260 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4262 (X86insrtps VR128:$src1,
4263 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4264 imm:$src3))]>, OpSize;
4267 let Constraints = "$src1 = $dst" in
4268 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4269 let Predicates = [HasAVX] in
4270 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4272 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4273 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4275 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4276 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4277 Requires<[HasSSE41]>;
4279 //===----------------------------------------------------------------------===//
4280 // SSE4.1 - Round Instructions
4281 //===----------------------------------------------------------------------===//
4283 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4284 X86MemOperand x86memop, RegisterClass RC,
4285 PatFrag mem_frag32, PatFrag mem_frag64,
4286 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4287 // Intrinsic operation, reg.
4288 // Vector intrinsic operation, reg
4289 def PSr : SS4AIi8<opcps, MRMSrcReg,
4290 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4291 !strconcat(OpcodeStr,
4292 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4293 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4296 // Vector intrinsic operation, mem
4297 def PSm : Ii8<opcps, MRMSrcMem,
4298 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4299 !strconcat(OpcodeStr,
4300 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4302 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4304 Requires<[HasSSE41]>;
4306 // Vector intrinsic operation, reg
4307 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4308 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4309 !strconcat(OpcodeStr,
4310 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4311 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4314 // Vector intrinsic operation, mem
4315 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4316 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4317 !strconcat(OpcodeStr,
4318 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4320 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4324 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4325 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4326 // Intrinsic operation, reg.
4327 // Vector intrinsic operation, reg
4328 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4329 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4330 !strconcat(OpcodeStr,
4331 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4334 // Vector intrinsic operation, mem
4335 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4336 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4337 !strconcat(OpcodeStr,
4338 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4339 []>, TA, OpSize, Requires<[HasSSE41]>;
4341 // Vector intrinsic operation, reg
4342 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4343 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4344 !strconcat(OpcodeStr,
4345 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4348 // Vector intrinsic operation, mem
4349 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4350 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4351 !strconcat(OpcodeStr,
4352 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4356 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4359 Intrinsic F64Int, bit Is2Addr = 1> {
4360 // Intrinsic operation, reg.
4361 def SSr : SS4AIi8<opcss, MRMSrcReg,
4362 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4364 !strconcat(OpcodeStr,
4365 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4366 !strconcat(OpcodeStr,
4367 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4368 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4371 // Intrinsic operation, mem.
4372 def SSm : SS4AIi8<opcss, MRMSrcMem,
4373 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4375 !strconcat(OpcodeStr,
4376 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4377 !strconcat(OpcodeStr,
4378 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4380 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4383 // Intrinsic operation, reg.
4384 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4385 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4387 !strconcat(OpcodeStr,
4388 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4389 !strconcat(OpcodeStr,
4390 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4391 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4394 // Intrinsic operation, mem.
4395 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4396 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4398 !strconcat(OpcodeStr,
4399 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4400 !strconcat(OpcodeStr,
4401 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4403 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4407 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4409 // Intrinsic operation, reg.
4410 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4411 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4412 !strconcat(OpcodeStr,
4413 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4416 // Intrinsic operation, mem.
4417 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4418 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4419 !strconcat(OpcodeStr,
4420 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4423 // Intrinsic operation, reg.
4424 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4425 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4426 !strconcat(OpcodeStr,
4427 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4430 // Intrinsic operation, mem.
4431 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4432 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4433 !strconcat(OpcodeStr,
4434 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4438 // FP round - roundss, roundps, roundsd, roundpd
4439 let Predicates = [HasAVX] in {
4441 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4442 memopv4f32, memopv2f64,
4443 int_x86_sse41_round_ps,
4444 int_x86_sse41_round_pd>, VEX;
4445 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4446 memopv8f32, memopv4f64,
4447 int_x86_avx_round_ps_256,
4448 int_x86_avx_round_pd_256>, VEX;
4449 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4450 int_x86_sse41_round_ss,
4451 int_x86_sse41_round_sd, 0>, VEX_4V;
4453 // Instructions for the assembler
4454 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4456 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4458 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4461 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4462 memopv4f32, memopv2f64,
4463 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4464 let Constraints = "$src1 = $dst" in
4465 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4466 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4468 //===----------------------------------------------------------------------===//
4469 // SSE4.1 - Packed Bit Test
4470 //===----------------------------------------------------------------------===//
4472 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4473 // the intel intrinsic that corresponds to this.
4474 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4475 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4476 "vptest\t{$src2, $src1|$src1, $src2}",
4477 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4479 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4480 "vptest\t{$src2, $src1|$src1, $src2}",
4481 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4484 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4485 "vptest\t{$src2, $src1|$src1, $src2}",
4486 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4488 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4489 "vptest\t{$src2, $src1|$src1, $src2}",
4490 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4494 let Defs = [EFLAGS] in {
4495 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4496 "ptest \t{$src2, $src1|$src1, $src2}",
4497 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4499 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4500 "ptest \t{$src2, $src1|$src1, $src2}",
4501 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4505 // The bit test instructions below are AVX only
4506 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4507 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4508 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4509 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4510 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4511 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4512 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4513 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4517 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4518 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4519 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4520 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4521 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4524 //===----------------------------------------------------------------------===//
4525 // SSE4.1 - Misc Instructions
4526 //===----------------------------------------------------------------------===//
4528 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4529 "popcnt{w}\t{$src, $dst|$dst, $src}",
4530 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4531 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4532 "popcnt{w}\t{$src, $dst|$dst, $src}",
4533 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4535 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4536 "popcnt{l}\t{$src, $dst|$dst, $src}",
4537 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4538 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4539 "popcnt{l}\t{$src, $dst|$dst, $src}",
4540 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4542 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4543 "popcnt{q}\t{$src, $dst|$dst, $src}",
4544 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4545 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4546 "popcnt{q}\t{$src, $dst|$dst, $src}",
4547 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4551 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4552 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4553 Intrinsic IntId128> {
4554 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4556 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4557 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4558 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4560 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4563 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4566 let Predicates = [HasAVX] in
4567 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4568 int_x86_sse41_phminposuw>, VEX;
4569 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4570 int_x86_sse41_phminposuw>;
4572 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4573 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4574 Intrinsic IntId128, bit Is2Addr = 1> {
4575 let isCommutable = 1 in
4576 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4577 (ins VR128:$src1, VR128:$src2),
4579 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4580 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4581 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4582 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4583 (ins VR128:$src1, i128mem:$src2),
4585 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4586 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4588 (IntId128 VR128:$src1,
4589 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4592 let Predicates = [HasAVX] in {
4593 let isCommutable = 0 in
4594 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4596 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4598 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4600 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4602 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4604 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4606 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4608 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4610 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4612 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4614 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4618 let Constraints = "$src1 = $dst" in {
4619 let isCommutable = 0 in
4620 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4621 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4622 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4623 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4624 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4625 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4626 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4627 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4628 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4629 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4630 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4633 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4634 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4635 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4636 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4638 /// SS48I_binop_rm - Simple SSE41 binary operator.
4639 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4640 ValueType OpVT, bit Is2Addr = 1> {
4641 let isCommutable = 1 in
4642 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4643 (ins VR128:$src1, VR128:$src2),
4645 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4646 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4647 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4649 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4650 (ins VR128:$src1, i128mem:$src2),
4652 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4653 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4654 [(set VR128:$dst, (OpNode VR128:$src1,
4655 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4659 let Predicates = [HasAVX] in
4660 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4661 let Constraints = "$src1 = $dst" in
4662 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4664 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4665 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4666 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4667 X86MemOperand x86memop, bit Is2Addr = 1> {
4668 let isCommutable = 1 in
4669 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4670 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4672 !strconcat(OpcodeStr,
4673 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4674 !strconcat(OpcodeStr,
4675 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4676 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4678 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4679 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4681 !strconcat(OpcodeStr,
4682 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4683 !strconcat(OpcodeStr,
4684 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4687 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4691 let Predicates = [HasAVX] in {
4692 let isCommutable = 0 in {
4693 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4694 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4695 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4696 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4697 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4698 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4699 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4700 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4701 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4702 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4703 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4704 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4706 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4707 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4708 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4709 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4710 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4711 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4714 let Constraints = "$src1 = $dst" in {
4715 let isCommutable = 0 in {
4716 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4717 VR128, memopv16i8, i128mem>;
4718 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4719 VR128, memopv16i8, i128mem>;
4720 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4721 VR128, memopv16i8, i128mem>;
4722 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4723 VR128, memopv16i8, i128mem>;
4725 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4726 VR128, memopv16i8, i128mem>;
4727 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4728 VR128, memopv16i8, i128mem>;
4731 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4732 let Predicates = [HasAVX] in {
4733 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4734 RegisterClass RC, X86MemOperand x86memop,
4735 PatFrag mem_frag, Intrinsic IntId> {
4736 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4737 (ins RC:$src1, RC:$src2, RC:$src3),
4738 !strconcat(OpcodeStr,
4739 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4740 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4741 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4743 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4744 (ins RC:$src1, x86memop:$src2, RC:$src3),
4745 !strconcat(OpcodeStr,
4746 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4748 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4750 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4754 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4755 memopv16i8, int_x86_sse41_blendvpd>;
4756 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4757 memopv16i8, int_x86_sse41_blendvps>;
4758 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4759 memopv16i8, int_x86_sse41_pblendvb>;
4760 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4761 memopv32i8, int_x86_avx_blendv_pd_256>;
4762 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4763 memopv32i8, int_x86_avx_blendv_ps_256>;
4765 /// SS41I_ternary_int - SSE 4.1 ternary operator
4766 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4767 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4768 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4769 (ins VR128:$src1, VR128:$src2),
4770 !strconcat(OpcodeStr,
4771 "\t{$src2, $dst|$dst, $src2}"),
4772 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4775 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4776 (ins VR128:$src1, i128mem:$src2),
4777 !strconcat(OpcodeStr,
4778 "\t{$src2, $dst|$dst, $src2}"),
4781 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4785 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4786 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4787 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4789 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4790 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4792 let Predicates = [HasAVX] in
4793 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4794 "vmovntdqa\t{$src, $dst|$dst, $src}",
4795 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4797 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4798 "movntdqa\t{$src, $dst|$dst, $src}",
4799 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4802 //===----------------------------------------------------------------------===//
4803 // SSE4.2 - Compare Instructions
4804 //===----------------------------------------------------------------------===//
4806 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4807 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4808 Intrinsic IntId128, bit Is2Addr = 1> {
4809 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4810 (ins VR128:$src1, VR128:$src2),
4812 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4813 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4814 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4816 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4817 (ins VR128:$src1, i128mem:$src2),
4819 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4820 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4822 (IntId128 VR128:$src1,
4823 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4826 let Predicates = [HasAVX] in
4827 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4829 let Constraints = "$src1 = $dst" in
4830 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4832 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4833 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4834 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4835 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4837 //===----------------------------------------------------------------------===//
4838 // SSE4.2 - String/text Processing Instructions
4839 //===----------------------------------------------------------------------===//
4841 // Packed Compare Implicit Length Strings, Return Mask
4842 multiclass pseudo_pcmpistrm<string asm> {
4843 def REG : PseudoI<(outs VR128:$dst),
4844 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4845 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4847 def MEM : PseudoI<(outs VR128:$dst),
4848 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4849 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4850 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4853 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4854 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4855 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4858 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4859 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4860 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4861 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4862 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4863 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4864 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4867 let Defs = [XMM0, EFLAGS] in {
4868 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4869 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4870 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4871 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4872 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4873 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4876 // Packed Compare Explicit Length Strings, Return Mask
4877 multiclass pseudo_pcmpestrm<string asm> {
4878 def REG : PseudoI<(outs VR128:$dst),
4879 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4880 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4881 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4882 def MEM : PseudoI<(outs VR128:$dst),
4883 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4884 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4885 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4888 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4889 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4890 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4893 let Predicates = [HasAVX],
4894 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4895 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4896 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4897 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4898 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4899 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4900 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4903 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4904 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4905 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4906 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4907 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4908 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4909 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4912 // Packed Compare Implicit Length Strings, Return Index
4913 let Defs = [ECX, EFLAGS] in {
4914 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4915 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4916 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4917 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4918 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4919 (implicit EFLAGS)]>, OpSize;
4920 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4921 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4922 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4923 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4924 (implicit EFLAGS)]>, OpSize;
4928 let Predicates = [HasAVX] in {
4929 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4931 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4933 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4935 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4937 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4939 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4943 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4944 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4945 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4946 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4947 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4948 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4950 // Packed Compare Explicit Length Strings, Return Index
4951 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4952 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4953 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4954 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4955 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4956 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4957 (implicit EFLAGS)]>, OpSize;
4958 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4959 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4960 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4962 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4963 (implicit EFLAGS)]>, OpSize;
4967 let Predicates = [HasAVX] in {
4968 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4970 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
4972 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
4974 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
4976 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
4978 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
4982 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
4983 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
4984 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
4985 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
4986 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
4987 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
4989 //===----------------------------------------------------------------------===//
4990 // SSE4.2 - CRC Instructions
4991 //===----------------------------------------------------------------------===//
4993 // No CRC instructions have AVX equivalents
4995 // crc intrinsic instruction
4996 // This set of instructions are only rm, the only difference is the size
4998 let Constraints = "$src1 = $dst" in {
4999 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5000 (ins GR32:$src1, i8mem:$src2),
5001 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5003 (int_x86_sse42_crc32_32_8 GR32:$src1,
5004 (load addr:$src2)))]>;
5005 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5006 (ins GR32:$src1, GR8:$src2),
5007 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5009 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5010 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5011 (ins GR32:$src1, i16mem:$src2),
5012 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5014 (int_x86_sse42_crc32_32_16 GR32:$src1,
5015 (load addr:$src2)))]>,
5017 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5018 (ins GR32:$src1, GR16:$src2),
5019 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5021 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5023 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5024 (ins GR32:$src1, i32mem:$src2),
5025 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5027 (int_x86_sse42_crc32_32_32 GR32:$src1,
5028 (load addr:$src2)))]>;
5029 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5030 (ins GR32:$src1, GR32:$src2),
5031 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5033 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5034 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5035 (ins GR64:$src1, i8mem:$src2),
5036 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5038 (int_x86_sse42_crc32_64_8 GR64:$src1,
5039 (load addr:$src2)))]>,
5041 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5042 (ins GR64:$src1, GR8:$src2),
5043 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5045 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5047 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5048 (ins GR64:$src1, i64mem:$src2),
5049 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5051 (int_x86_sse42_crc32_64_64 GR64:$src1,
5052 (load addr:$src2)))]>,
5054 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5055 (ins GR64:$src1, GR64:$src2),
5056 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5058 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5062 //===----------------------------------------------------------------------===//
5063 // AES-NI Instructions
5064 //===----------------------------------------------------------------------===//
5066 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5067 Intrinsic IntId128, bit Is2Addr = 1> {
5068 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5069 (ins VR128:$src1, VR128:$src2),
5071 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5072 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5073 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5075 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5076 (ins VR128:$src1, i128mem:$src2),
5078 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5079 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5081 (IntId128 VR128:$src1,
5082 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5085 // Perform One Round of an AES Encryption/Decryption Flow
5086 let Predicates = [HasAVX, HasAES] in {
5087 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5088 int_x86_aesni_aesenc, 0>, VEX_4V;
5089 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5090 int_x86_aesni_aesenclast, 0>, VEX_4V;
5091 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5092 int_x86_aesni_aesdec, 0>, VEX_4V;
5093 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5094 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5097 let Constraints = "$src1 = $dst" in {
5098 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5099 int_x86_aesni_aesenc>;
5100 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5101 int_x86_aesni_aesenclast>;
5102 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5103 int_x86_aesni_aesdec>;
5104 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5105 int_x86_aesni_aesdeclast>;
5108 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5109 (AESENCrr VR128:$src1, VR128:$src2)>;
5110 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5111 (AESENCrm VR128:$src1, addr:$src2)>;
5112 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5113 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5114 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5115 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5116 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5117 (AESDECrr VR128:$src1, VR128:$src2)>;
5118 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5119 (AESDECrm VR128:$src1, addr:$src2)>;
5120 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5121 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5122 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5123 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5125 // Perform the AES InvMixColumn Transformation
5126 let Predicates = [HasAVX, HasAES] in {
5127 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5129 "vaesimc\t{$src1, $dst|$dst, $src1}",
5131 (int_x86_aesni_aesimc VR128:$src1))]>,
5133 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5134 (ins i128mem:$src1),
5135 "vaesimc\t{$src1, $dst|$dst, $src1}",
5137 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5140 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5142 "aesimc\t{$src1, $dst|$dst, $src1}",
5144 (int_x86_aesni_aesimc VR128:$src1))]>,
5146 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5147 (ins i128mem:$src1),
5148 "aesimc\t{$src1, $dst|$dst, $src1}",
5150 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5153 // AES Round Key Generation Assist
5154 let Predicates = [HasAVX, HasAES] in {
5155 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5156 (ins VR128:$src1, i8imm:$src2),
5157 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5159 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5161 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5162 (ins i128mem:$src1, i8imm:$src2),
5163 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5165 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5169 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5170 (ins VR128:$src1, i8imm:$src2),
5171 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5173 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5175 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5176 (ins i128mem:$src1, i8imm:$src2),
5177 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5179 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5183 //===----------------------------------------------------------------------===//
5184 // CLMUL Instructions
5185 //===----------------------------------------------------------------------===//
5187 // Carry-less Multiplication instructions
5188 let Constraints = "$src1 = $dst" in {
5189 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5190 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5191 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5194 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5195 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5196 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5200 // AVX carry-less Multiplication instructions
5201 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5202 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5203 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5206 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5207 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5208 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5212 multiclass pclmul_alias<string asm, int immop> {
5213 def : InstAlias<!strconcat("pclmul", asm,
5214 "dq {$src, $dst|$dst, $src}"),
5215 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5217 def : InstAlias<!strconcat("pclmul", asm,
5218 "dq {$src, $dst|$dst, $src}"),
5219 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5221 def : InstAlias<!strconcat("vpclmul", asm,
5222 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5223 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5225 def : InstAlias<!strconcat("vpclmul", asm,
5226 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5227 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5229 defm : pclmul_alias<"hqhq", 0x11>;
5230 defm : pclmul_alias<"hqlq", 0x01>;
5231 defm : pclmul_alias<"lqhq", 0x10>;
5232 defm : pclmul_alias<"lqlq", 0x00>;
5234 //===----------------------------------------------------------------------===//
5236 //===----------------------------------------------------------------------===//
5239 // Load from memory and broadcast to all elements of the destination operand
5240 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5241 X86MemOperand x86memop, Intrinsic Int> :
5242 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5243 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5244 [(set RC:$dst, (Int addr:$src))]>, VEX;
5246 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5247 int_x86_avx_vbroadcastss>;
5248 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5249 int_x86_avx_vbroadcastss_256>;
5250 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5251 int_x86_avx_vbroadcast_sd_256>;
5252 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5253 int_x86_avx_vbroadcastf128_pd_256>;
5255 // Insert packed floating-point values
5256 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5257 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5258 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5260 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5261 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5262 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5265 // Extract packed floating-point values
5266 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5267 (ins VR256:$src1, i8imm:$src2),
5268 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5270 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5271 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5272 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5275 // Conditional SIMD Packed Loads and Stores
5276 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5277 Intrinsic IntLd, Intrinsic IntLd256,
5278 Intrinsic IntSt, Intrinsic IntSt256,
5279 PatFrag pf128, PatFrag pf256> {
5280 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5281 (ins VR128:$src1, f128mem:$src2),
5282 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5283 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5285 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5286 (ins VR256:$src1, f256mem:$src2),
5287 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5288 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5290 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5291 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5292 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5293 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5294 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5295 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5296 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5297 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5300 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5301 int_x86_avx_maskload_ps,
5302 int_x86_avx_maskload_ps_256,
5303 int_x86_avx_maskstore_ps,
5304 int_x86_avx_maskstore_ps_256,
5305 memopv4f32, memopv8f32>;
5306 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5307 int_x86_avx_maskload_pd,
5308 int_x86_avx_maskload_pd_256,
5309 int_x86_avx_maskstore_pd,
5310 int_x86_avx_maskstore_pd_256,
5311 memopv2f64, memopv4f64>;
5313 // Permute Floating-Point Values
5314 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5315 RegisterClass RC, X86MemOperand x86memop_f,
5316 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5317 Intrinsic IntVar, Intrinsic IntImm> {
5318 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5319 (ins RC:$src1, RC:$src2),
5320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5321 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5322 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5323 (ins RC:$src1, x86memop_i:$src2),
5324 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5325 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5327 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5328 (ins RC:$src1, i8imm:$src2),
5329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5330 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5331 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5332 (ins x86memop_f:$src1, i8imm:$src2),
5333 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5334 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5337 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5338 memopv4f32, memopv4i32,
5339 int_x86_avx_vpermilvar_ps,
5340 int_x86_avx_vpermil_ps>;
5341 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5342 memopv8f32, memopv8i32,
5343 int_x86_avx_vpermilvar_ps_256,
5344 int_x86_avx_vpermil_ps_256>;
5345 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5346 memopv2f64, memopv2i64,
5347 int_x86_avx_vpermilvar_pd,
5348 int_x86_avx_vpermil_pd>;
5349 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5350 memopv4f64, memopv4i64,
5351 int_x86_avx_vpermilvar_pd_256,
5352 int_x86_avx_vpermil_pd_256>;
5354 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5355 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5356 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5358 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5359 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5360 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5363 // Zero All YMM registers
5364 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5365 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5367 // Zero Upper bits of YMM registers
5368 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5369 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5371 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5372 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5373 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5374 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5375 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5376 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5378 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5380 (VINSERTF128rr VR256:$src1, VR128:$src2,
5381 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5382 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5384 (VINSERTF128rr VR256:$src1, VR128:$src2,
5385 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5386 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5388 (VINSERTF128rr VR256:$src1, VR128:$src2,
5389 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5390 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5392 (VINSERTF128rr VR256:$src1, VR128:$src2,
5393 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5395 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5396 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5397 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5398 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5399 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5400 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5402 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5403 (v4f32 (VEXTRACTF128rr
5404 (v8f32 VR256:$src1),
5405 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5406 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5407 (v2f64 (VEXTRACTF128rr
5408 (v4f64 VR256:$src1),
5409 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5410 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5411 (v4i32 (VEXTRACTF128rr
5412 (v8i32 VR256:$src1),
5413 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5414 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5415 (v2i64 (VEXTRACTF128rr
5416 (v4i64 VR256:$src1),
5417 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5419 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5420 (VBROADCASTF128 addr:$src)>;
5422 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5423 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5424 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5425 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5426 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5427 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5429 def : Pat<(int_x86_avx_vperm2f128_ps_256
5430 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5431 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5432 def : Pat<(int_x86_avx_vperm2f128_pd_256
5433 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5434 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5435 def : Pat<(int_x86_avx_vperm2f128_si_256
5436 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5437 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5439 //===----------------------------------------------------------------------===//
5440 // SSE Shuffle pattern fragments
5441 //===----------------------------------------------------------------------===//
5443 // This is part of a "work in progress" refactoring. The idea is that all
5444 // vector shuffles are going to be translated into target specific nodes and
5445 // directly matched by the patterns below (which can be changed along the way)
5446 // The AVX version of some but not all of them are described here, and more
5447 // should come in a near future.
5449 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5450 // SSE2 loads, which are always promoted to v2i64. The last one should match
5451 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5452 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5453 // we investigate further.
5454 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5456 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5457 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5459 (PSHUFDmi addr:$src1, imm:$imm)>;
5460 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5462 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5464 // Shuffle with PSHUFD instruction.
5465 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5466 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5467 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5468 (PSHUFDri VR128:$src1, imm:$imm)>;
5470 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5471 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5472 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5473 (PSHUFDri VR128:$src1, imm:$imm)>;
5475 // Shuffle with SHUFPD instruction.
5476 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5477 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5478 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5479 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5480 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5481 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5483 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5484 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5485 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5486 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5488 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5489 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5490 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5491 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5493 // Shuffle with SHUFPS instruction.
5494 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5495 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5496 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5497 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5498 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5499 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5501 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5502 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5503 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5504 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5506 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5507 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5508 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5509 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5510 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5511 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5513 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5514 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5515 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5516 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5518 // Shuffle with MOVHLPS instruction
5519 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5520 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5521 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5522 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5524 // Shuffle with MOVDDUP instruction
5525 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5526 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5527 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5528 (MOVDDUPrm addr:$src)>;
5530 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5531 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5532 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5533 (MOVDDUPrm addr:$src)>;
5535 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5536 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5537 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5538 (MOVDDUPrm addr:$src)>;
5540 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5541 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5542 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5543 (MOVDDUPrm addr:$src)>;
5545 def : Pat<(X86Movddup (bc_v2f64
5546 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5547 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5548 def : Pat<(X86Movddup (bc_v2f64
5549 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5550 (MOVDDUPrm addr:$src)>;
5553 // Shuffle with UNPCKLPS
5554 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5555 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5556 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5557 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5558 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5559 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5561 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5562 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5563 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5564 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5565 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5566 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5568 // Shuffle with UNPCKHPS
5569 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5570 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5571 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5572 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5574 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5575 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5576 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5577 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5579 // Shuffle with UNPCKLPD
5580 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5581 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5582 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5583 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5584 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5585 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5587 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5588 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5589 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5590 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5591 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5592 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5594 // Shuffle with UNPCKHPD
5595 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5596 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5597 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5598 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5600 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5601 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5602 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5603 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5605 // Shuffle with PUNPCKLBW
5606 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5607 (bc_v16i8 (memopv2i64 addr:$src2)))),
5608 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5609 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5610 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5612 // Shuffle with PUNPCKLWD
5613 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5614 (bc_v8i16 (memopv2i64 addr:$src2)))),
5615 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5616 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5617 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5619 // Shuffle with PUNPCKLDQ
5620 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5621 (bc_v4i32 (memopv2i64 addr:$src2)))),
5622 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5623 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5624 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5626 // Shuffle with PUNPCKLQDQ
5627 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5628 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5629 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5630 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5632 // Shuffle with PUNPCKHBW
5633 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5634 (bc_v16i8 (memopv2i64 addr:$src2)))),
5635 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5636 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5637 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5639 // Shuffle with PUNPCKHWD
5640 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5641 (bc_v8i16 (memopv2i64 addr:$src2)))),
5642 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5643 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5644 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5646 // Shuffle with PUNPCKHDQ
5647 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5648 (bc_v4i32 (memopv2i64 addr:$src2)))),
5649 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5650 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5651 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5653 // Shuffle with PUNPCKHQDQ
5654 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5655 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5656 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5657 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5659 // Shuffle with MOVLHPS
5660 def : Pat<(X86Movlhps VR128:$src1,
5661 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5662 (MOVHPSrm VR128:$src1, addr:$src2)>;
5663 def : Pat<(X86Movlhps VR128:$src1,
5664 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5665 (MOVHPSrm VR128:$src1, addr:$src2)>;
5666 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5667 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5668 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5669 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5670 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5671 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5673 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5674 // is during lowering, where it's not possible to recognize the load fold cause
5675 // it has two uses through a bitcast. One use disappears at isel time and the
5676 // fold opportunity reappears.
5677 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5678 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5680 // Shuffle with MOVLHPD
5681 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5682 (scalar_to_vector (loadf64 addr:$src2)))),
5683 (MOVHPDrm VR128:$src1, addr:$src2)>;
5685 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5686 // is during lowering, where it's not possible to recognize the load fold cause
5687 // it has two uses through a bitcast. One use disappears at isel time and the
5688 // fold opportunity reappears.
5689 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5690 (scalar_to_vector (loadf64 addr:$src2)))),
5691 (MOVHPDrm VR128:$src1, addr:$src2)>;
5693 // Shuffle with MOVSS
5694 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5695 (MOVSSrr VR128:$src1, FR32:$src2)>;
5696 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5697 (MOVSSrr (v4i32 VR128:$src1),
5698 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5699 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5700 (MOVSSrr (v4f32 VR128:$src1),
5701 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5702 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5703 // is during lowering, where it's not possible to recognize the load fold cause
5704 // it has two uses through a bitcast. One use disappears at isel time and the
5705 // fold opportunity reappears.
5706 def : Pat<(X86Movss VR128:$src1,
5707 (bc_v4i32 (v2i64 (load addr:$src2)))),
5708 (MOVLPSrm VR128:$src1, addr:$src2)>;
5710 // Shuffle with MOVSD
5711 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5712 (MOVSDrr VR128:$src1, FR64:$src2)>;
5713 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5714 (MOVSDrr (v2i64 VR128:$src1),
5715 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5716 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5717 (MOVSDrr (v2f64 VR128:$src1),
5718 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5719 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5720 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5721 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5722 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5724 // Shuffle with MOVSHDUP
5725 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5726 (MOVSHDUPrr VR128:$src)>;
5727 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5728 (MOVSHDUPrm addr:$src)>;
5730 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5731 (MOVSHDUPrr VR128:$src)>;
5732 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5733 (MOVSHDUPrm addr:$src)>;
5735 // Shuffle with MOVSLDUP
5736 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5737 (MOVSLDUPrr VR128:$src)>;
5738 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5739 (MOVSLDUPrm addr:$src)>;
5741 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5742 (MOVSLDUPrr VR128:$src)>;
5743 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5744 (MOVSLDUPrm addr:$src)>;
5746 // Shuffle with PSHUFHW
5747 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5748 (PSHUFHWri VR128:$src, imm:$imm)>;
5749 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5750 (PSHUFHWmi addr:$src, imm:$imm)>;
5752 // Shuffle with PSHUFLW
5753 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5754 (PSHUFLWri VR128:$src, imm:$imm)>;
5755 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5756 (PSHUFLWmi addr:$src, imm:$imm)>;
5758 // Shuffle with PALIGN
5759 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5760 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5761 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5762 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5763 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5764 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5765 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5766 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5768 // Shuffle with MOVLPS
5769 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5770 (MOVLPSrm VR128:$src1, addr:$src2)>;
5771 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5772 (MOVLPSrm VR128:$src1, addr:$src2)>;
5773 def : Pat<(X86Movlps VR128:$src1,
5774 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5775 (MOVLPSrm VR128:$src1, addr:$src2)>;
5776 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5777 // is during lowering, where it's not possible to recognize the load fold cause
5778 // it has two uses through a bitcast. One use disappears at isel time and the
5779 // fold opportunity reappears.
5780 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5781 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5783 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5784 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5786 // Shuffle with MOVLPD
5787 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5788 (MOVLPDrm VR128:$src1, addr:$src2)>;
5789 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5790 (MOVLPDrm VR128:$src1, addr:$src2)>;
5791 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5792 (scalar_to_vector (loadf64 addr:$src2)))),
5793 (MOVLPDrm VR128:$src1, addr:$src2)>;
5795 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5796 def : Pat<(store (f64 (vector_extract
5797 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5798 (MOVHPSmr addr:$dst, VR128:$src)>;
5799 def : Pat<(store (f64 (vector_extract
5800 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5801 (MOVHPDmr addr:$dst, VR128:$src)>;
5803 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5804 (MOVLPSmr addr:$src1, VR128:$src2)>;
5805 def : Pat<(store (v4i32 (X86Movlps
5806 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5807 (MOVLPSmr addr:$src1, VR128:$src2)>;
5809 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5810 (MOVLPDmr addr:$src1, VR128:$src2)>;
5811 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5812 (MOVLPDmr addr:$src1, VR128:$src2)>;