1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // SSE 1 & 2 - Move Instructions
120 //===----------------------------------------------------------------------===//
122 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
123 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
124 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
126 // Loading from memory automatically zeroing upper bits.
127 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
128 PatFrag mem_pat, string OpcodeStr> :
129 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
131 [(set RC:$dst, (mem_pat addr:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
139 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
140 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
141 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
143 let canFoldAsLoad = 1, isReMaterializable = 1 in {
144 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
146 let AddedComplexity = 20 in
147 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
150 let Constraints = "$src1 = $dst" in {
151 def MOVSSrr : sse12_move_rr<FR32, v4f32,
152 "movss\t{$src2, $dst|$dst, $src2}">, XS;
153 def MOVSDrr : sse12_move_rr<FR64, v2f64,
154 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
157 let canFoldAsLoad = 1, isReMaterializable = 1 in {
158 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
160 let AddedComplexity = 20 in
161 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
164 let AddedComplexity = 15 in {
165 // Extract the low 32-bit value from one vector and insert it into another.
166 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
167 (MOVSSrr (v4f32 VR128:$src1),
168 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
169 // Extract the low 64-bit value from one vector and insert it into another.
170 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
171 (MOVSDrr (v2f64 VR128:$src1),
172 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
175 // Implicitly promote a 32-bit scalar to a vector.
176 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
177 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
178 // Implicitly promote a 64-bit scalar to a vector.
179 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
180 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
181 // Implicitly promote a 32-bit scalar to a vector.
182 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
183 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
184 // Implicitly promote a 64-bit scalar to a vector.
185 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
186 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
188 let AddedComplexity = 20 in {
189 // MOVSSrm zeros the high parts of the register; represent this
190 // with SUBREG_TO_REG.
191 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
192 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
193 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 // MOVSDrm zeros the high parts of the register; represent this
198 // with SUBREG_TO_REG.
199 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
200 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
201 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzload addr:$src)),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
211 // Store scalar value to memory.
212 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
213 "movss\t{$src, $dst|$dst, $src}",
214 [(store FR32:$src, addr:$dst)]>;
215 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
216 "movsd\t{$src, $dst|$dst, $src}",
217 [(store FR64:$src, addr:$dst)]>;
219 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
220 "movss\t{$src, $dst|$dst, $src}",
221 [(store FR32:$src, addr:$dst)]>, XS, VEX;
222 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
223 "movsd\t{$src, $dst|$dst, $src}",
224 [(store FR64:$src, addr:$dst)]>, XD, VEX;
226 // Extract and store.
227 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
230 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
231 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
236 // Move Aligned/Unaligned floating point values
237 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
238 X86MemOperand x86memop, PatFrag ld_frag,
239 string asm, Domain d,
240 bit IsReMaterializable = 1> {
241 let neverHasSideEffects = 1 in
242 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
243 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
244 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
245 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
246 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
247 [(set RC:$dst, (ld_frag addr:$src))], d>;
250 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
251 "movaps", SSEPackedSingle>, VEX;
252 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
253 "movapd", SSEPackedDouble>, OpSize, VEX;
254 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
255 "movups", SSEPackedSingle>, VEX;
256 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
257 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
259 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
260 "movaps", SSEPackedSingle>, VEX;
261 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
262 "movapd", SSEPackedDouble>, OpSize, VEX;
263 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
264 "movups", SSEPackedSingle>, VEX;
265 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
266 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
267 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
268 "movaps", SSEPackedSingle>, TB;
269 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
270 "movapd", SSEPackedDouble>, TB, OpSize;
271 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
272 "movups", SSEPackedSingle>, TB;
273 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
274 "movupd", SSEPackedDouble, 0>, TB, OpSize;
276 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
277 "movaps\t{$src, $dst|$dst, $src}",
278 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
279 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
280 "movapd\t{$src, $dst|$dst, $src}",
281 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
282 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
283 "movups\t{$src, $dst|$dst, $src}",
284 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
285 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
286 "movupd\t{$src, $dst|$dst, $src}",
287 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
288 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
289 "movaps\t{$src, $dst|$dst, $src}",
290 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
291 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
292 "movapd\t{$src, $dst|$dst, $src}",
293 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
294 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
295 "movups\t{$src, $dst|$dst, $src}",
296 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
297 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
298 "movupd\t{$src, $dst|$dst, $src}",
299 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
302 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
303 (VMOVUPSYmr addr:$dst, VR256:$src)>;
305 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
306 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
307 (VMOVUPDYmr addr:$dst, VR256:$src)>;
309 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
310 "movaps\t{$src, $dst|$dst, $src}",
311 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
312 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
313 "movapd\t{$src, $dst|$dst, $src}",
314 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
315 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
316 "movups\t{$src, $dst|$dst, $src}",
317 [(store (v4f32 VR128:$src), addr:$dst)]>;
318 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
319 "movupd\t{$src, $dst|$dst, $src}",
320 [(store (v2f64 VR128:$src), addr:$dst)]>;
322 // Intrinsic forms of MOVUPS/D load and store
323 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
324 (ins f128mem:$dst, VR128:$src),
325 "movups\t{$src, $dst|$dst, $src}",
326 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
327 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
328 (ins f128mem:$dst, VR128:$src),
329 "movupd\t{$src, $dst|$dst, $src}",
330 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
332 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
333 "movups\t{$src, $dst|$dst, $src}",
334 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
335 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
336 "movupd\t{$src, $dst|$dst, $src}",
337 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
339 // Move Low/High packed floating point values
340 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
341 PatFrag mov_frag, string base_opc,
343 def PSrm : PI<opc, MRMSrcMem,
344 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
345 !strconcat(base_opc, "s", asm_opr),
348 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
349 SSEPackedSingle>, TB;
351 def PDrm : PI<opc, MRMSrcMem,
352 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
353 !strconcat(base_opc, "d", asm_opr),
354 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
355 (scalar_to_vector (loadf64 addr:$src2)))))],
356 SSEPackedDouble>, TB, OpSize;
359 let AddedComplexity = 20 in {
360 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
361 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
362 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
363 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
365 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
366 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
367 "\t{$src2, $dst|$dst, $src2}">;
368 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
369 "\t{$src2, $dst|$dst, $src2}">;
372 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
373 "movlps\t{$src, $dst|$dst, $src}",
374 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
375 (iPTR 0))), addr:$dst)]>, VEX;
376 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
377 "movlpd\t{$src, $dst|$dst, $src}",
378 [(store (f64 (vector_extract (v2f64 VR128:$src),
379 (iPTR 0))), addr:$dst)]>, VEX;
380 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
381 "movlps\t{$src, $dst|$dst, $src}",
382 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
383 (iPTR 0))), addr:$dst)]>;
384 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
385 "movlpd\t{$src, $dst|$dst, $src}",
386 [(store (f64 (vector_extract (v2f64 VR128:$src),
387 (iPTR 0))), addr:$dst)]>;
389 // v2f64 extract element 1 is always custom lowered to unpack high to low
390 // and extract element 0 so the non-store version isn't too horrible.
391 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
392 "movhps\t{$src, $dst|$dst, $src}",
393 [(store (f64 (vector_extract
394 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
395 (undef)), (iPTR 0))), addr:$dst)]>,
397 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
398 "movhpd\t{$src, $dst|$dst, $src}",
399 [(store (f64 (vector_extract
400 (v2f64 (unpckh VR128:$src, (undef))),
401 (iPTR 0))), addr:$dst)]>,
403 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movhps\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract
406 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
407 (undef)), (iPTR 0))), addr:$dst)]>;
408 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movhpd\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract
411 (v2f64 (unpckh VR128:$src, (undef))),
412 (iPTR 0))), addr:$dst)]>;
414 let AddedComplexity = 20 in {
415 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
416 (ins VR128:$src1, VR128:$src2),
417 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
419 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
421 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
422 (ins VR128:$src1, VR128:$src2),
423 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
425 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
428 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
429 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
430 (ins VR128:$src1, VR128:$src2),
431 "movlhps\t{$src2, $dst|$dst, $src2}",
433 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
434 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
435 (ins VR128:$src1, VR128:$src2),
436 "movhlps\t{$src2, $dst|$dst, $src2}",
438 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
441 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
442 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
443 let AddedComplexity = 20 in {
444 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
445 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
446 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
447 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
450 //===----------------------------------------------------------------------===//
451 // SSE 1 & 2 - Conversion Instructions
452 //===----------------------------------------------------------------------===//
454 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
455 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
457 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
458 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
459 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
460 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
463 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
464 X86MemOperand x86memop, string asm> {
465 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
467 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
471 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
472 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
473 string asm, Domain d> {
474 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
475 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
476 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
477 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
480 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
481 X86MemOperand x86memop, string asm> {
482 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
483 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
484 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
485 (ins DstRC:$src1, x86memop:$src),
486 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
489 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
490 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
491 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
492 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
494 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
495 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
496 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
497 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
500 // The assembler can recognize rr 64-bit instructions by seeing a rxx
501 // register, but the same isn't true when only using memory operands,
502 // provide other assembly "l" and "q" forms to address this explicitly
503 // where appropriate to do so.
504 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
506 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
508 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
510 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
512 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
515 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
516 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
517 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
518 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
519 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
520 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
521 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
522 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
523 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
524 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
525 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
526 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
527 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
528 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
529 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
530 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
532 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
533 // and/or XMM operand(s).
535 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
536 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
538 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
539 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
540 [(set DstRC:$dst, (Int SrcRC:$src))]>;
541 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
542 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
543 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
546 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
547 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
548 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
549 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
551 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
552 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
553 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
554 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
555 (ins DstRC:$src1, x86memop:$src2),
557 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
558 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
559 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
562 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
563 f32mem, load, "cvtss2si">, XS, VEX;
564 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
565 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
567 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
568 f128mem, load, "cvtsd2si">, XD, VEX;
569 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
570 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
573 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
574 // Get rid of this hack or rename the intrinsics, there are several
575 // intructions that only match with the intrinsic form, why create duplicates
576 // to let them be recognized by the assembler?
577 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
578 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
579 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
580 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
581 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
582 f32mem, load, "cvtss2si">, XS;
583 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
584 f32mem, load, "cvtss2si{q}">, XS, REX_W;
585 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
586 f128mem, load, "cvtsd2si{l}">, XD;
587 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
588 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
591 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
592 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
593 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
594 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
596 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
597 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
598 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
599 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
602 let Constraints = "$src1 = $dst" in {
603 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
604 int_x86_sse_cvtsi2ss, i32mem, loadi32,
606 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
607 int_x86_sse_cvtsi642ss, i64mem, loadi64,
608 "cvtsi2ss{q}">, XS, REX_W;
609 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
610 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
612 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
613 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
614 "cvtsi2sd">, XD, REX_W;
619 // Aliases for intrinsics
620 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
621 f32mem, load, "cvttss2si">, XS, VEX;
622 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
623 int_x86_sse_cvttss2si64, f32mem, load,
624 "cvttss2si">, XS, VEX, VEX_W;
625 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
626 f128mem, load, "cvttsd2si">, XD, VEX;
627 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
628 int_x86_sse2_cvttsd2si64, f128mem, load,
629 "cvttsd2si">, XD, VEX, VEX_W;
630 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
631 f32mem, load, "cvttss2si">, XS;
632 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
633 int_x86_sse_cvttss2si64, f32mem, load,
634 "cvttss2si{q}">, XS, REX_W;
635 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
636 f128mem, load, "cvttsd2si">, XD;
637 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
638 int_x86_sse2_cvttsd2si64, f128mem, load,
639 "cvttsd2si{q}">, XD, REX_W;
641 let Pattern = []<dag> in {
642 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
643 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
644 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
645 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
647 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
648 "cvtdq2ps\t{$src, $dst|$dst, $src}",
649 SSEPackedSingle>, TB, VEX;
650 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
651 "cvtdq2ps\t{$src, $dst|$dst, $src}",
652 SSEPackedSingle>, TB, VEX;
654 let Pattern = []<dag> in {
655 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
656 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
657 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
658 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
659 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
660 "cvtdq2ps\t{$src, $dst|$dst, $src}",
661 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
666 // Convert scalar double to scalar single
667 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
668 (ins FR64:$src1, FR64:$src2),
669 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
671 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
672 (ins FR64:$src1, f64mem:$src2),
673 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
674 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
675 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
678 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
679 "cvtsd2ss\t{$src, $dst|$dst, $src}",
680 [(set FR32:$dst, (fround FR64:$src))]>;
681 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
682 "cvtsd2ss\t{$src, $dst|$dst, $src}",
683 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
684 Requires<[HasSSE2, OptForSize]>;
686 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
687 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
689 let Constraints = "$src1 = $dst" in
690 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
691 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
693 // Convert scalar single to scalar double
694 // SSE2 instructions with XS prefix
695 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
696 (ins FR32:$src1, FR32:$src2),
697 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
698 []>, XS, Requires<[HasAVX]>, VEX_4V;
699 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
700 (ins FR32:$src1, f32mem:$src2),
701 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
702 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
703 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
706 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
707 "cvtss2sd\t{$src, $dst|$dst, $src}",
708 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
710 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
711 "cvtss2sd\t{$src, $dst|$dst, $src}",
712 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
713 Requires<[HasSSE2, OptForSize]>;
715 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
717 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
718 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
719 VR128:$src2))]>, XS, VEX_4V,
721 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
722 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
725 (load addr:$src2)))]>, XS, VEX_4V,
727 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
728 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
729 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
730 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
731 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
734 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
735 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
738 (load addr:$src2)))]>, XS,
742 def : Pat<(extloadf32 addr:$src),
743 (CVTSS2SDrr (MOVSSrm addr:$src))>,
744 Requires<[HasSSE2, OptForSpeed]>;
746 // Convert doubleword to packed single/double fp
747 // SSE2 instructions without OpSize prefix
748 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
749 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
750 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
751 TB, VEX, Requires<[HasAVX]>;
752 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
753 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
754 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
755 (bitconvert (memopv2i64 addr:$src))))]>,
756 TB, VEX, Requires<[HasAVX]>;
757 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
758 "cvtdq2ps\t{$src, $dst|$dst, $src}",
759 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
760 TB, Requires<[HasSSE2]>;
761 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
762 "cvtdq2ps\t{$src, $dst|$dst, $src}",
763 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
764 (bitconvert (memopv2i64 addr:$src))))]>,
765 TB, Requires<[HasSSE2]>;
767 // FIXME: why the non-intrinsic version is described as SSE3?
768 // SSE2 instructions with XS prefix
769 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
770 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
771 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
772 XS, VEX, Requires<[HasAVX]>;
773 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
774 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
775 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
776 (bitconvert (memopv2i64 addr:$src))))]>,
777 XS, VEX, Requires<[HasAVX]>;
778 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
779 "cvtdq2pd\t{$src, $dst|$dst, $src}",
780 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
781 XS, Requires<[HasSSE2]>;
782 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
783 "cvtdq2pd\t{$src, $dst|$dst, $src}",
784 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
785 (bitconvert (memopv2i64 addr:$src))))]>,
786 XS, Requires<[HasSSE2]>;
789 // Convert packed single/double fp to doubleword
790 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
791 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
792 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
794 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
796 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
798 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
800 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
803 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
804 "cvtps2dq\t{$src, $dst|$dst, $src}",
805 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
807 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
809 "cvtps2dq\t{$src, $dst|$dst, $src}",
810 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
811 (memop addr:$src)))]>, VEX;
812 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
813 "cvtps2dq\t{$src, $dst|$dst, $src}",
814 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
815 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
816 "cvtps2dq\t{$src, $dst|$dst, $src}",
817 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
818 (memop addr:$src)))]>;
820 // SSE2 packed instructions with XD prefix
821 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
822 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
823 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
824 XD, VEX, Requires<[HasAVX]>;
825 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
826 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
827 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
828 (memop addr:$src)))]>,
829 XD, VEX, Requires<[HasAVX]>;
830 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
831 "cvtpd2dq\t{$src, $dst|$dst, $src}",
832 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
833 XD, Requires<[HasSSE2]>;
834 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
835 "cvtpd2dq\t{$src, $dst|$dst, $src}",
836 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
837 (memop addr:$src)))]>,
838 XD, Requires<[HasSSE2]>;
841 // Convert with truncation packed single/double fp to doubleword
842 // SSE2 packed instructions with XS prefix
843 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
844 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
845 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
847 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
848 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
849 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
850 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
851 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
852 "cvttps2dq\t{$src, $dst|$dst, $src}",
854 (int_x86_sse2_cvttps2dq VR128:$src))]>;
855 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
856 "cvttps2dq\t{$src, $dst|$dst, $src}",
858 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
861 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
862 "vcvttps2dq\t{$src, $dst|$dst, $src}",
864 (int_x86_sse2_cvttps2dq VR128:$src))]>,
865 XS, VEX, Requires<[HasAVX]>;
866 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
867 "vcvttps2dq\t{$src, $dst|$dst, $src}",
868 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
869 (memop addr:$src)))]>,
870 XS, VEX, Requires<[HasAVX]>;
872 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
874 "cvttpd2dq\t{$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
877 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
879 "cvttpd2dq\t{$src, $dst|$dst, $src}",
880 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
881 (memop addr:$src)))]>, VEX;
882 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
883 "cvttpd2dq\t{$src, $dst|$dst, $src}",
884 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
885 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
886 "cvttpd2dq\t{$src, $dst|$dst, $src}",
887 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
888 (memop addr:$src)))]>;
890 // The assembler can recognize rr 256-bit instructions by seeing a ymm
891 // register, but the same isn't true when using memory operands instead.
892 // Provide other assembly rr and rm forms to address this explicitly.
893 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
894 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
895 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
896 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
899 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
900 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
901 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
902 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
905 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
906 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
907 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
908 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
910 // Convert packed single to packed double
911 let Predicates = [HasAVX] in {
912 // SSE2 instructions without OpSize prefix
913 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
914 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
915 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
916 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
917 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
918 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
919 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
920 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
922 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
923 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
924 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
925 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
927 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
928 "vcvtps2pd\t{$src, $dst|$dst, $src}",
929 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
930 VEX, Requires<[HasAVX]>;
931 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
932 "vcvtps2pd\t{$src, $dst|$dst, $src}",
933 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
934 (load addr:$src)))]>,
935 VEX, Requires<[HasAVX]>;
936 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
937 "cvtps2pd\t{$src, $dst|$dst, $src}",
938 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
939 TB, Requires<[HasSSE2]>;
940 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
941 "cvtps2pd\t{$src, $dst|$dst, $src}",
942 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
943 (load addr:$src)))]>,
944 TB, Requires<[HasSSE2]>;
946 // Convert packed double to packed single
947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
948 // register, but the same isn't true when using memory operands instead.
949 // Provide other assembly rr and rm forms to address this explicitly.
950 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
952 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
953 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
963 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
964 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
965 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
966 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
967 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
968 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
969 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
972 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
973 "cvtpd2ps\t{$src, $dst|$dst, $src}",
974 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
975 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
977 "cvtpd2ps\t{$src, $dst|$dst, $src}",
978 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
979 (memop addr:$src)))]>;
980 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
981 "cvtpd2ps\t{$src, $dst|$dst, $src}",
982 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
983 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
984 "cvtpd2ps\t{$src, $dst|$dst, $src}",
985 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
986 (memop addr:$src)))]>;
988 // AVX 256-bit register conversion intrinsics
989 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
990 // whenever possible to avoid declaring two versions of each one.
991 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
992 (VCVTDQ2PSYrr VR256:$src)>;
993 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
994 (VCVTDQ2PSYrm addr:$src)>;
996 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
997 (VCVTPD2PSYrr VR256:$src)>;
998 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
999 (VCVTPD2PSYrm addr:$src)>;
1001 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1002 (VCVTPS2DQYrr VR256:$src)>;
1003 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1004 (VCVTPS2DQYrm addr:$src)>;
1006 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1007 (VCVTPS2PDYrr VR128:$src)>;
1008 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1009 (VCVTPS2PDYrm addr:$src)>;
1011 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1012 (VCVTTPD2DQYrr VR256:$src)>;
1013 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1014 (VCVTTPD2DQYrm addr:$src)>;
1016 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1017 (VCVTTPS2DQYrr VR256:$src)>;
1018 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1019 (VCVTTPS2DQYrm addr:$src)>;
1021 //===----------------------------------------------------------------------===//
1022 // SSE 1 & 2 - Compare Instructions
1023 //===----------------------------------------------------------------------===//
1025 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1026 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1027 string asm, string asm_alt> {
1028 let isAsmParserOnly = 1 in {
1029 def rr : SIi8<0xC2, MRMSrcReg,
1030 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1033 def rm : SIi8<0xC2, MRMSrcMem,
1034 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1038 // Accept explicit immediate argument form instead of comparison code.
1039 def rr_alt : SIi8<0xC2, MRMSrcReg,
1040 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1043 def rm_alt : SIi8<0xC2, MRMSrcMem,
1044 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1048 let neverHasSideEffects = 1 in {
1049 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1050 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1051 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1053 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1054 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1055 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1059 let Constraints = "$src1 = $dst" in {
1060 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1061 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1062 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1063 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1064 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1065 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1066 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1067 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1068 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1069 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1070 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1071 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1072 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1073 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1074 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1075 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1077 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1078 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1079 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1080 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1081 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1082 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1083 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1084 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1085 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1086 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1087 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1088 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1089 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1092 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1093 Intrinsic Int, string asm> {
1094 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1095 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1096 [(set VR128:$dst, (Int VR128:$src1,
1097 VR128:$src, imm:$cc))]>;
1098 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1099 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1100 [(set VR128:$dst, (Int VR128:$src1,
1101 (load addr:$src), imm:$cc))]>;
1104 // Aliases to match intrinsics which expect XMM operand(s).
1105 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1106 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1108 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1109 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1111 let Constraints = "$src1 = $dst" in {
1112 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1113 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1114 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1115 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1119 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1120 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1121 ValueType vt, X86MemOperand x86memop,
1122 PatFrag ld_frag, string OpcodeStr, Domain d> {
1123 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1124 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1125 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1126 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1127 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1128 [(set EFLAGS, (OpNode (vt RC:$src1),
1129 (ld_frag addr:$src2)))], d>;
1132 let Defs = [EFLAGS] in {
1133 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1134 "ucomiss", SSEPackedSingle>, VEX;
1135 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1136 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1137 let Pattern = []<dag> in {
1138 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1139 "comiss", SSEPackedSingle>, VEX;
1140 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1141 "comisd", SSEPackedDouble>, OpSize, VEX;
1144 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1145 load, "ucomiss", SSEPackedSingle>, VEX;
1146 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1147 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1149 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1150 load, "comiss", SSEPackedSingle>, VEX;
1151 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1152 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1153 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1154 "ucomiss", SSEPackedSingle>, TB;
1155 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1156 "ucomisd", SSEPackedDouble>, TB, OpSize;
1158 let Pattern = []<dag> in {
1159 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1160 "comiss", SSEPackedSingle>, TB;
1161 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1162 "comisd", SSEPackedDouble>, TB, OpSize;
1165 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1166 load, "ucomiss", SSEPackedSingle>, TB;
1167 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1168 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1170 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1171 "comiss", SSEPackedSingle>, TB;
1172 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1173 "comisd", SSEPackedDouble>, TB, OpSize;
1174 } // Defs = [EFLAGS]
1176 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1177 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1178 Intrinsic Int, string asm, string asm_alt,
1180 let isAsmParserOnly = 1 in {
1181 def rri : PIi8<0xC2, MRMSrcReg,
1182 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1183 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1184 def rmi : PIi8<0xC2, MRMSrcMem,
1185 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1186 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1189 // Accept explicit immediate argument form instead of comparison code.
1190 def rri_alt : PIi8<0xC2, MRMSrcReg,
1191 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1193 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1194 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1198 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1199 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1200 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1201 SSEPackedSingle>, VEX_4V;
1202 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1203 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1204 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1205 SSEPackedDouble>, OpSize, VEX_4V;
1206 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1207 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1208 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1209 SSEPackedSingle>, VEX_4V;
1210 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1211 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1212 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1213 SSEPackedDouble>, OpSize, VEX_4V;
1214 let Constraints = "$src1 = $dst" in {
1215 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1216 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1217 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1218 SSEPackedSingle>, TB;
1219 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1220 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1221 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1222 SSEPackedDouble>, TB, OpSize;
1225 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1226 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1227 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1228 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1229 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1230 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1231 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1232 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1234 //===----------------------------------------------------------------------===//
1235 // SSE 1 & 2 - Shuffle Instructions
1236 //===----------------------------------------------------------------------===//
1238 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1239 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1240 ValueType vt, string asm, PatFrag mem_frag,
1241 Domain d, bit IsConvertibleToThreeAddress = 0> {
1242 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1243 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1244 [(set RC:$dst, (vt (shufp:$src3
1245 RC:$src1, (mem_frag addr:$src2))))], d>;
1246 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1247 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1248 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1250 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1253 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1254 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1255 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1256 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1257 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1258 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1259 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1260 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1261 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1262 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1263 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1264 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1266 let Constraints = "$src1 = $dst" in {
1267 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1268 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1269 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1271 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1272 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1273 memopv2f64, SSEPackedDouble>, TB, OpSize;
1276 //===----------------------------------------------------------------------===//
1277 // SSE 1 & 2 - Unpack Instructions
1278 //===----------------------------------------------------------------------===//
1280 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1281 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1282 PatFrag mem_frag, RegisterClass RC,
1283 X86MemOperand x86memop, string asm,
1285 def rr : PI<opc, MRMSrcReg,
1286 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1288 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1289 def rm : PI<opc, MRMSrcMem,
1290 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1292 (vt (OpNode RC:$src1,
1293 (mem_frag addr:$src2))))], d>;
1296 let AddedComplexity = 10 in {
1297 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1298 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1299 SSEPackedSingle>, VEX_4V;
1300 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1301 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1302 SSEPackedDouble>, OpSize, VEX_4V;
1303 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1304 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1305 SSEPackedSingle>, VEX_4V;
1306 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1307 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1308 SSEPackedDouble>, OpSize, VEX_4V;
1310 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1311 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1312 SSEPackedSingle>, VEX_4V;
1313 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1314 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1315 SSEPackedDouble>, OpSize, VEX_4V;
1316 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1317 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1318 SSEPackedSingle>, VEX_4V;
1319 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1320 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1321 SSEPackedDouble>, OpSize, VEX_4V;
1323 let Constraints = "$src1 = $dst" in {
1324 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1325 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1326 SSEPackedSingle>, TB;
1327 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1328 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1329 SSEPackedDouble>, TB, OpSize;
1330 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1331 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1332 SSEPackedSingle>, TB;
1333 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1334 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1335 SSEPackedDouble>, TB, OpSize;
1336 } // Constraints = "$src1 = $dst"
1337 } // AddedComplexity
1339 //===----------------------------------------------------------------------===//
1340 // SSE 1 & 2 - Extract Floating-Point Sign mask
1341 //===----------------------------------------------------------------------===//
1343 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1344 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1346 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1347 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1348 [(set GR32:$dst, (Int RC:$src))], d>;
1349 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1350 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1354 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1355 "movmskps", SSEPackedSingle>, VEX;
1356 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1357 "movmskpd", SSEPackedDouble>, OpSize,
1359 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1360 "movmskps", SSEPackedSingle>, VEX;
1361 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1362 "movmskpd", SSEPackedDouble>, OpSize,
1364 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1365 SSEPackedSingle>, TB;
1366 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1367 SSEPackedDouble>, TB, OpSize;
1370 def MOVMSKPDrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR64:$src),
1371 "movmskpd\t{$src, $dst|$dst, $src}",
1372 [(set GR32:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1373 def MOVMSKPDrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR64:$src),
1374 "movmskpd\t{$src, $dst|$dst, $src}",
1375 [(set GR64:$dst, (X86fgetsign FR64:$src))], SSEPackedDouble>, TB, OpSize;
1376 def MOVMSKPSrr32_alt : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins FR32:$src),
1377 "movmskps\t{$src, $dst|$dst, $src}",
1378 [(set GR32:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1379 def MOVMSKPSrr64_alt : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins FR32:$src),
1380 "movmskps\t{$src, $dst|$dst, $src}",
1381 [(set GR64:$dst, (X86fgetsign FR32:$src))], SSEPackedSingle>, TB;
1384 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1385 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1386 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1387 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1389 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1390 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1391 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1392 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1395 //===----------------------------------------------------------------------===//
1396 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1397 //===----------------------------------------------------------------------===//
1399 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1400 // names that start with 'Fs'.
1402 // Alias instructions that map fld0 to pxor for sse.
1403 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1404 canFoldAsLoad = 1 in {
1405 // FIXME: Set encoding to pseudo!
1406 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1407 [(set FR32:$dst, fp32imm0)]>,
1408 Requires<[HasSSE1]>, TB, OpSize;
1409 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1410 [(set FR64:$dst, fpimm0)]>,
1411 Requires<[HasSSE2]>, TB, OpSize;
1412 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1413 [(set FR32:$dst, fp32imm0)]>,
1414 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1415 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1416 [(set FR64:$dst, fpimm0)]>,
1417 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1420 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1421 // bits are disregarded.
1422 let neverHasSideEffects = 1 in {
1423 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1424 "movaps\t{$src, $dst|$dst, $src}", []>;
1425 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1426 "movapd\t{$src, $dst|$dst, $src}", []>;
1429 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1430 // bits are disregarded.
1431 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1432 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1433 "movaps\t{$src, $dst|$dst, $src}",
1434 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1435 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1436 "movapd\t{$src, $dst|$dst, $src}",
1437 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1440 //===----------------------------------------------------------------------===//
1441 // SSE 1 & 2 - Logical Instructions
1442 //===----------------------------------------------------------------------===//
1444 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1446 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1448 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1449 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1451 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1452 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1454 let Constraints = "$src1 = $dst" in {
1455 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1456 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1458 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1459 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1463 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1464 let mayLoad = 0 in {
1465 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1466 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1467 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1470 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1471 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1473 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1475 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1476 SDNode OpNode, int HasPat = 0,
1477 list<list<dag>> Pattern = []> {
1478 let Pattern = []<dag> in {
1479 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1480 !strconcat(OpcodeStr, "ps"), f128mem,
1481 !if(HasPat, Pattern[0], // rr
1482 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1484 !if(HasPat, Pattern[2], // rm
1485 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1486 (memopv2i64 addr:$src2)))]), 0>,
1489 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1490 !strconcat(OpcodeStr, "pd"), f128mem,
1491 !if(HasPat, Pattern[1], // rr
1492 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1495 !if(HasPat, Pattern[3], // rm
1496 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1497 (memopv2i64 addr:$src2)))]), 0>,
1500 let Constraints = "$src1 = $dst" in {
1501 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1502 !strconcat(OpcodeStr, "ps"), f128mem,
1503 !if(HasPat, Pattern[0], // rr
1504 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1506 !if(HasPat, Pattern[2], // rm
1507 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1508 (memopv2i64 addr:$src2)))])>, TB;
1510 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1511 !strconcat(OpcodeStr, "pd"), f128mem,
1512 !if(HasPat, Pattern[1], // rr
1513 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1516 !if(HasPat, Pattern[3], // rm
1517 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1518 (memopv2i64 addr:$src2)))])>,
1523 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1525 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1526 SDNode OpNode, int HasNoPat = 0> {
1527 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1528 !strconcat(OpcodeStr, "ps"), f256mem,
1529 !if(HasNoPat, []<dag>, // rr
1530 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1,
1532 !if(HasNoPat, []<dag>, // rm
1533 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1534 (memopv4i64 addr:$src2)))]), 0>, VEX_4V;
1536 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1537 !strconcat(OpcodeStr, "pd"), f256mem,
1538 !if(HasNoPat, []<dag>, // rr
1539 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1540 (bc_v4i64 (v4f64 VR256:$src2))))]),
1541 !if(HasNoPat, []<dag>, // rm
1542 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1543 (memopv4i64 addr:$src2)))]), 0>,
1547 // AVX 256-bit packed logical ops forms
1548 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1549 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1550 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1551 let isCommutable = 0 in {
1552 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", undef /* dummy */, 1>;
1555 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1556 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1557 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1558 let isCommutable = 0 in
1559 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1561 [(set VR128:$dst, (X86andnp VR128:$src1, VR128:$src2))],
1565 [(set VR128:$dst, (X86andnp VR128:$src1, (memopv2i64 addr:$src2)))],
1569 //===----------------------------------------------------------------------===//
1570 // SSE 1 & 2 - Arithmetic Instructions
1571 //===----------------------------------------------------------------------===//
1573 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1576 /// In addition, we also have a special variant of the scalar form here to
1577 /// represent the associated intrinsic operation. This form is unlike the
1578 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1579 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1581 /// These three forms can each be reg+reg or reg+mem.
1584 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1586 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1588 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1589 OpNode, FR32, f32mem, Is2Addr>, XS;
1590 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1591 OpNode, FR64, f64mem, Is2Addr>, XD;
1594 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1596 let mayLoad = 0 in {
1597 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1598 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1599 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1600 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1604 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1606 let mayLoad = 0 in {
1607 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1608 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1609 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1610 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1614 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1616 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1617 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1618 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1619 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1622 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1624 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1625 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1626 SSEPackedSingle, Is2Addr>, TB;
1628 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1629 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1630 SSEPackedDouble, Is2Addr>, TB, OpSize;
1633 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1634 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1635 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1636 SSEPackedSingle, 0>, TB;
1638 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1639 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1640 SSEPackedDouble, 0>, TB, OpSize;
1643 // Binary Arithmetic instructions
1644 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1645 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1646 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1647 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1648 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1649 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1650 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1651 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1653 let isCommutable = 0 in {
1654 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1655 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1656 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1657 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1658 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1659 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1660 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1661 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1662 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1663 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1664 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1665 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1666 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1667 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1668 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1669 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1670 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1671 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1672 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1673 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1676 let Constraints = "$src1 = $dst" in {
1677 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1678 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1679 basic_sse12_fp_binop_s_int<0x58, "add">;
1680 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1681 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1682 basic_sse12_fp_binop_s_int<0x59, "mul">;
1684 let isCommutable = 0 in {
1685 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1686 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1687 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1688 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1689 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1690 basic_sse12_fp_binop_s_int<0x5E, "div">;
1691 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1692 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1693 basic_sse12_fp_binop_s_int<0x5F, "max">,
1694 basic_sse12_fp_binop_p_int<0x5F, "max">;
1695 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1696 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1697 basic_sse12_fp_binop_s_int<0x5D, "min">,
1698 basic_sse12_fp_binop_p_int<0x5D, "min">;
1703 /// In addition, we also have a special variant of the scalar form here to
1704 /// represent the associated intrinsic operation. This form is unlike the
1705 /// plain scalar form, in that it takes an entire vector (instead of a
1706 /// scalar) and leaves the top elements undefined.
1708 /// And, we have a special variant form for a full-vector intrinsic form.
1710 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1711 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1712 SDNode OpNode, Intrinsic F32Int> {
1713 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1714 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1715 [(set FR32:$dst, (OpNode FR32:$src))]>;
1716 // For scalar unary operations, fold a load into the operation
1717 // only in OptForSize mode. It eliminates an instruction, but it also
1718 // eliminates a whole-register clobber (the load), so it introduces a
1719 // partial register update condition.
1720 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1721 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1722 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1723 Requires<[HasSSE1, OptForSize]>;
1724 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1725 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1726 [(set VR128:$dst, (F32Int VR128:$src))]>;
1727 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1728 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1729 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1732 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1733 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1734 SDNode OpNode, Intrinsic F32Int> {
1735 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1736 !strconcat(OpcodeStr,
1737 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1738 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1739 !strconcat(OpcodeStr,
1740 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1741 []>, XS, Requires<[HasAVX, OptForSize]>;
1742 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1743 !strconcat(OpcodeStr,
1744 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1745 [(set VR128:$dst, (F32Int VR128:$src))]>;
1746 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1747 !strconcat(OpcodeStr,
1748 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1749 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1752 /// sse1_fp_unop_p - SSE1 unops in packed form.
1753 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1754 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1755 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1756 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1757 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1758 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1759 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1762 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1763 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1764 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1765 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1766 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1767 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1768 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1769 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1772 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1773 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1774 Intrinsic V4F32Int> {
1775 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1777 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1778 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1779 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1780 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1783 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1784 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1785 Intrinsic V4F32Int> {
1786 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1787 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1788 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1789 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1790 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1791 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1794 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1795 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1796 SDNode OpNode, Intrinsic F64Int> {
1797 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1798 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1799 [(set FR64:$dst, (OpNode FR64:$src))]>;
1800 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1801 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1802 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1803 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1804 Requires<[HasSSE2, OptForSize]>;
1805 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1806 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1807 [(set VR128:$dst, (F64Int VR128:$src))]>;
1808 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1809 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1810 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1813 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1814 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1815 SDNode OpNode, Intrinsic F64Int> {
1816 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1817 !strconcat(OpcodeStr,
1818 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1819 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1820 (ins FR64:$src1, f64mem:$src2),
1821 !strconcat(OpcodeStr,
1822 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1823 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1825 [(set VR128:$dst, (F64Int VR128:$src))]>;
1826 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1827 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1828 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1831 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1832 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1834 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1835 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1836 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1837 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1838 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1839 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1842 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1843 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1844 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1845 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1846 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1847 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1848 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1849 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1852 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1853 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1854 Intrinsic V2F64Int> {
1855 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1856 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1857 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1858 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1859 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1860 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1863 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1864 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1865 Intrinsic V2F64Int> {
1866 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1867 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1868 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1869 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1870 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1871 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1874 let Predicates = [HasAVX] in {
1876 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1877 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1880 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1881 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1882 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1883 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1884 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1885 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1886 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1887 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1890 // Reciprocal approximations. Note that these typically require refinement
1891 // in order to obtain suitable precision.
1892 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1893 int_x86_sse_rsqrt_ss>, VEX_4V;
1894 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1895 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1896 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1897 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1899 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1901 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1902 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1903 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1904 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1908 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1909 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1910 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1911 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1912 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1913 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1915 // Reciprocal approximations. Note that these typically require refinement
1916 // in order to obtain suitable precision.
1917 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1918 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1919 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1920 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1921 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1922 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1924 // There is no f64 version of the reciprocal approximation instructions.
1926 //===----------------------------------------------------------------------===//
1927 // SSE 1 & 2 - Non-temporal stores
1928 //===----------------------------------------------------------------------===//
1930 let AddedComplexity = 400 in { // Prefer non-temporal versions
1931 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1932 (ins f128mem:$dst, VR128:$src),
1933 "movntps\t{$src, $dst|$dst, $src}",
1934 [(alignednontemporalstore (v4f32 VR128:$src),
1936 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1937 (ins f128mem:$dst, VR128:$src),
1938 "movntpd\t{$src, $dst|$dst, $src}",
1939 [(alignednontemporalstore (v2f64 VR128:$src),
1941 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1942 (ins f128mem:$dst, VR128:$src),
1943 "movntdq\t{$src, $dst|$dst, $src}",
1944 [(alignednontemporalstore (v2f64 VR128:$src),
1947 let ExeDomain = SSEPackedInt in
1948 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1949 (ins f128mem:$dst, VR128:$src),
1950 "movntdq\t{$src, $dst|$dst, $src}",
1951 [(alignednontemporalstore (v4f32 VR128:$src),
1954 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
1955 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
1957 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1958 (ins f256mem:$dst, VR256:$src),
1959 "movntps\t{$src, $dst|$dst, $src}",
1960 [(alignednontemporalstore (v8f32 VR256:$src),
1962 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
1963 (ins f256mem:$dst, VR256:$src),
1964 "movntpd\t{$src, $dst|$dst, $src}",
1965 [(alignednontemporalstore (v4f64 VR256:$src),
1967 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
1968 (ins f256mem:$dst, VR256:$src),
1969 "movntdq\t{$src, $dst|$dst, $src}",
1970 [(alignednontemporalstore (v4f64 VR256:$src),
1972 let ExeDomain = SSEPackedInt in
1973 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
1974 (ins f256mem:$dst, VR256:$src),
1975 "movntdq\t{$src, $dst|$dst, $src}",
1976 [(alignednontemporalstore (v8f32 VR256:$src),
1980 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
1981 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
1982 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
1983 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
1984 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
1985 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
1987 let AddedComplexity = 400 in { // Prefer non-temporal versions
1988 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1989 "movntps\t{$src, $dst|$dst, $src}",
1990 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
1991 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1992 "movntpd\t{$src, $dst|$dst, $src}",
1993 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
1995 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
1996 "movntdq\t{$src, $dst|$dst, $src}",
1997 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
1999 let ExeDomain = SSEPackedInt in
2000 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2001 "movntdq\t{$src, $dst|$dst, $src}",
2002 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2004 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2005 (MOVNTDQmr addr:$dst, VR128:$src)>;
2007 // There is no AVX form for instructions below this point
2008 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2009 "movnti{l}\t{$src, $dst|$dst, $src}",
2010 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2011 TB, Requires<[HasSSE2]>;
2012 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2013 "movnti{q}\t{$src, $dst|$dst, $src}",
2014 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2015 TB, Requires<[HasSSE2]>;
2018 //===----------------------------------------------------------------------===//
2019 // SSE 1 & 2 - Misc Instructions (No AVX form)
2020 //===----------------------------------------------------------------------===//
2022 // Prefetch intrinsic.
2023 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2024 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2025 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2026 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2027 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2028 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2029 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2030 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2032 // Load, store, and memory fence
2033 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2034 TB, Requires<[HasSSE1]>;
2035 def : Pat<(X86SFence), (SFENCE)>;
2037 // Alias instructions that map zero vector to pxor / xorp* for sse.
2038 // We set canFoldAsLoad because this can be converted to a constant-pool
2039 // load of an all-zeros value if folding it would be beneficial.
2040 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2041 // JIT implementation, it does not expand the instructions below like
2042 // X86MCInstLower does.
2043 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2044 isCodeGenOnly = 1 in {
2045 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2046 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2047 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2048 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2049 let ExeDomain = SSEPackedInt in
2050 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2051 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2054 // The same as done above but for AVX. The 128-bit versions are the
2055 // same, but re-encoded. The 256-bit does not support PI version.
2056 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2057 // JIT implementatioan, it does not expand the instructions below like
2058 // X86MCInstLower does.
2059 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2060 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2061 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2062 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2063 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2064 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2065 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2066 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2067 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2068 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2069 let ExeDomain = SSEPackedInt in
2070 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2071 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2074 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2075 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2076 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2078 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2079 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2081 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
2082 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
2083 // represent this instead of always zeroing SRC1. One possible solution is
2084 // to represent the instruction w/ something similar as the "$src1 = $dst"
2085 // constraint but without the tied operands.
2086 def : Pat<(extloadf32 addr:$src),
2087 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)), addr:$src)>,
2088 Requires<[HasAVX, OptForSpeed]>;
2090 //===----------------------------------------------------------------------===//
2091 // SSE 1 & 2 - Load/Store XCSR register
2092 //===----------------------------------------------------------------------===//
2094 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2095 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2096 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2097 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2099 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2100 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2101 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2102 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2104 //===---------------------------------------------------------------------===//
2105 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2106 //===---------------------------------------------------------------------===//
2108 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2110 let neverHasSideEffects = 1 in {
2111 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2112 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2113 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2114 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2116 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2117 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2118 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2119 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2121 let canFoldAsLoad = 1, mayLoad = 1 in {
2122 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2123 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2124 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2125 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2126 let Predicates = [HasAVX] in {
2127 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2128 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2129 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2130 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2134 let mayStore = 1 in {
2135 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2136 (ins i128mem:$dst, VR128:$src),
2137 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2138 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2139 (ins i256mem:$dst, VR256:$src),
2140 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2141 let Predicates = [HasAVX] in {
2142 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2143 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2144 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2145 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2149 let neverHasSideEffects = 1 in
2150 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2151 "movdqa\t{$src, $dst|$dst, $src}", []>;
2153 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2154 "movdqu\t{$src, $dst|$dst, $src}",
2155 []>, XS, Requires<[HasSSE2]>;
2157 let canFoldAsLoad = 1, mayLoad = 1 in {
2158 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2159 "movdqa\t{$src, $dst|$dst, $src}",
2160 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2161 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2162 "movdqu\t{$src, $dst|$dst, $src}",
2163 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2164 XS, Requires<[HasSSE2]>;
2167 let mayStore = 1 in {
2168 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2169 "movdqa\t{$src, $dst|$dst, $src}",
2170 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2171 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2172 "movdqu\t{$src, $dst|$dst, $src}",
2173 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2174 XS, Requires<[HasSSE2]>;
2177 // Intrinsic forms of MOVDQU load and store
2178 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2179 "vmovdqu\t{$src, $dst|$dst, $src}",
2180 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2181 XS, VEX, Requires<[HasAVX]>;
2183 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2184 "movdqu\t{$src, $dst|$dst, $src}",
2185 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2186 XS, Requires<[HasSSE2]>;
2188 } // ExeDomain = SSEPackedInt
2190 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2191 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2192 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2194 //===---------------------------------------------------------------------===//
2195 // SSE2 - Packed Integer Arithmetic Instructions
2196 //===---------------------------------------------------------------------===//
2198 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2200 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2201 bit IsCommutable = 0, bit Is2Addr = 1> {
2202 let isCommutable = IsCommutable in
2203 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2204 (ins VR128:$src1, VR128:$src2),
2206 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2207 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2208 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2209 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2210 (ins VR128:$src1, i128mem:$src2),
2212 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2213 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2214 [(set VR128:$dst, (IntId VR128:$src1,
2215 (bitconvert (memopv2i64 addr:$src2))))]>;
2218 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2219 string OpcodeStr, Intrinsic IntId,
2220 Intrinsic IntId2, bit Is2Addr = 1> {
2221 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2222 (ins VR128:$src1, VR128:$src2),
2224 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2225 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2226 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2227 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2228 (ins VR128:$src1, i128mem:$src2),
2230 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2231 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2232 [(set VR128:$dst, (IntId VR128:$src1,
2233 (bitconvert (memopv2i64 addr:$src2))))]>;
2234 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2235 (ins VR128:$src1, i32i8imm:$src2),
2237 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2239 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2242 /// PDI_binop_rm - Simple SSE2 binary operator.
2243 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2244 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2245 let isCommutable = IsCommutable in
2246 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2247 (ins VR128:$src1, VR128:$src2),
2249 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2250 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2251 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2252 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2253 (ins VR128:$src1, i128mem:$src2),
2255 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2256 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2257 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2258 (bitconvert (memopv2i64 addr:$src2)))))]>;
2261 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2263 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2264 /// to collapse (bitconvert VT to VT) into its operand.
2266 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2267 bit IsCommutable = 0, bit Is2Addr = 1> {
2268 let isCommutable = IsCommutable in
2269 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2270 (ins VR128:$src1, VR128:$src2),
2272 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2273 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2274 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2275 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2276 (ins VR128:$src1, i128mem:$src2),
2278 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2279 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2280 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2283 } // ExeDomain = SSEPackedInt
2285 // 128-bit Integer Arithmetic
2287 let Predicates = [HasAVX] in {
2288 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2289 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2290 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2291 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2292 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2293 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2294 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2295 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2296 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2299 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2301 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2303 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2305 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2307 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2309 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2311 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2313 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2315 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2317 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2319 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2321 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2323 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2325 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2327 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2329 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2331 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2333 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2335 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2339 let Constraints = "$src1 = $dst" in {
2340 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2341 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2342 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2343 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2344 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2345 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2346 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2347 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2348 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2351 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2352 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2353 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2354 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2355 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2356 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2357 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2358 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2359 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2360 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2361 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2362 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2363 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2364 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2365 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2366 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2367 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2368 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2369 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2371 } // Constraints = "$src1 = $dst"
2373 //===---------------------------------------------------------------------===//
2374 // SSE2 - Packed Integer Logical Instructions
2375 //===---------------------------------------------------------------------===//
2377 let Predicates = [HasAVX] in {
2378 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2379 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2381 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2382 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2384 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2385 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2388 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2389 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2391 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2392 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2394 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2395 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2398 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2399 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2401 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2402 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2405 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2406 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2407 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2409 let ExeDomain = SSEPackedInt in {
2410 let neverHasSideEffects = 1 in {
2411 // 128-bit logical shifts.
2412 def VPSLLDQri : PDIi8<0x73, MRM7r,
2413 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2414 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2416 def VPSRLDQri : PDIi8<0x73, MRM3r,
2417 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2418 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2420 // PSRADQri doesn't exist in SSE[1-3].
2422 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2423 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2424 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2425 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2426 VR128:$src2)))]>, VEX_4V;
2428 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2429 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2430 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2431 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2432 (memopv2i64 addr:$src2))))]>,
2437 let Constraints = "$src1 = $dst" in {
2438 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2439 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2440 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2441 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2442 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2443 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2445 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2446 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2447 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2448 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2449 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2450 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2452 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2453 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2454 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2455 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2457 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2458 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2459 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2461 let ExeDomain = SSEPackedInt in {
2462 let neverHasSideEffects = 1 in {
2463 // 128-bit logical shifts.
2464 def PSLLDQri : PDIi8<0x73, MRM7r,
2465 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2466 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2467 def PSRLDQri : PDIi8<0x73, MRM3r,
2468 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2469 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2470 // PSRADQri doesn't exist in SSE[1-3].
2472 def PANDNrr : PDI<0xDF, MRMSrcReg,
2473 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2474 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2476 def PANDNrm : PDI<0xDF, MRMSrcMem,
2477 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2478 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2480 } // Constraints = "$src1 = $dst"
2482 let Predicates = [HasAVX] in {
2483 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2484 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2485 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2486 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2487 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2488 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2489 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2490 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2491 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2492 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2494 // Shift up / down and insert zero's.
2495 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2496 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2497 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2498 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2501 let Predicates = [HasSSE2] in {
2502 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2503 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2504 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2505 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2506 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2507 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2508 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2509 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2510 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2511 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2513 // Shift up / down and insert zero's.
2514 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2515 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2516 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2517 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2520 //===---------------------------------------------------------------------===//
2521 // SSE2 - Packed Integer Comparison Instructions
2522 //===---------------------------------------------------------------------===//
2524 let Predicates = [HasAVX] in {
2525 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2527 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2529 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2531 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2533 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2535 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2539 let Constraints = "$src1 = $dst" in {
2540 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2541 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2542 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2543 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2544 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2545 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2546 } // Constraints = "$src1 = $dst"
2548 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2549 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2550 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2551 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2552 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2553 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2554 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2555 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2556 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2557 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2558 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2559 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2561 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2562 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2563 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2564 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2565 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2566 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2567 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2568 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2569 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2570 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2571 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2572 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2574 //===---------------------------------------------------------------------===//
2575 // SSE2 - Packed Integer Pack Instructions
2576 //===---------------------------------------------------------------------===//
2578 let Predicates = [HasAVX] in {
2579 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2581 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2583 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2587 let Constraints = "$src1 = $dst" in {
2588 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2589 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2590 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2591 } // Constraints = "$src1 = $dst"
2593 //===---------------------------------------------------------------------===//
2594 // SSE2 - Packed Integer Shuffle Instructions
2595 //===---------------------------------------------------------------------===//
2597 let ExeDomain = SSEPackedInt in {
2598 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2600 def ri : Ii8<0x70, MRMSrcReg,
2601 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2602 !strconcat(OpcodeStr,
2603 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2604 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2606 def mi : Ii8<0x70, MRMSrcMem,
2607 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2608 !strconcat(OpcodeStr,
2609 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2610 [(set VR128:$dst, (vt (pshuf_frag:$src2
2611 (bc_frag (memopv2i64 addr:$src1)),
2614 } // ExeDomain = SSEPackedInt
2616 let Predicates = [HasAVX] in {
2617 let AddedComplexity = 5 in
2618 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2621 // SSE2 with ImmT == Imm8 and XS prefix.
2622 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2625 // SSE2 with ImmT == Imm8 and XD prefix.
2626 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2630 let Predicates = [HasSSE2] in {
2631 let AddedComplexity = 5 in
2632 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2634 // SSE2 with ImmT == Imm8 and XS prefix.
2635 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2637 // SSE2 with ImmT == Imm8 and XD prefix.
2638 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2641 //===---------------------------------------------------------------------===//
2642 // SSE2 - Packed Integer Unpack Instructions
2643 //===---------------------------------------------------------------------===//
2645 let ExeDomain = SSEPackedInt in {
2646 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2647 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2648 def rr : PDI<opc, MRMSrcReg,
2649 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2651 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2652 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2653 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2654 def rm : PDI<opc, MRMSrcMem,
2655 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2657 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2658 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2659 [(set VR128:$dst, (unp_frag VR128:$src1,
2660 (bc_frag (memopv2i64
2664 let Predicates = [HasAVX] in {
2665 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2667 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2669 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2672 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2673 /// knew to collapse (bitconvert VT to VT) into its operand.
2674 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2675 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2676 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2678 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2679 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2680 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2681 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2683 (v2i64 (unpckl VR128:$src1,
2684 (memopv2i64 addr:$src2))))]>, VEX_4V;
2686 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2688 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2690 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2693 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2694 /// knew to collapse (bitconvert VT to VT) into its operand.
2695 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2696 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2697 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2699 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2700 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2701 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2702 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2704 (v2i64 (unpckh VR128:$src1,
2705 (memopv2i64 addr:$src2))))]>, VEX_4V;
2708 let Constraints = "$src1 = $dst" in {
2709 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2710 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2711 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2713 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2714 /// knew to collapse (bitconvert VT to VT) into its operand.
2715 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2717 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2719 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2720 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2721 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2722 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2724 (v2i64 (unpckl VR128:$src1,
2725 (memopv2i64 addr:$src2))))]>;
2727 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2728 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2729 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2731 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2732 /// knew to collapse (bitconvert VT to VT) into its operand.
2733 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2734 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2735 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2737 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2738 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2739 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2740 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2742 (v2i64 (unpckh VR128:$src1,
2743 (memopv2i64 addr:$src2))))]>;
2746 } // ExeDomain = SSEPackedInt
2748 //===---------------------------------------------------------------------===//
2749 // SSE2 - Packed Integer Extract and Insert
2750 //===---------------------------------------------------------------------===//
2752 let ExeDomain = SSEPackedInt in {
2753 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2754 def rri : Ii8<0xC4, MRMSrcReg,
2755 (outs VR128:$dst), (ins VR128:$src1,
2756 GR32:$src2, i32i8imm:$src3),
2758 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2759 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2761 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2762 def rmi : Ii8<0xC4, MRMSrcMem,
2763 (outs VR128:$dst), (ins VR128:$src1,
2764 i16mem:$src2, i32i8imm:$src3),
2766 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2767 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2769 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2774 let Predicates = [HasAVX] in
2775 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2776 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2777 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2778 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2779 imm:$src2))]>, OpSize, VEX;
2780 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2781 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2782 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2783 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2787 let Predicates = [HasAVX] in {
2788 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2789 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2790 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2791 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2792 []>, OpSize, VEX_4V;
2795 let Constraints = "$src1 = $dst" in
2796 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2798 } // ExeDomain = SSEPackedInt
2800 //===---------------------------------------------------------------------===//
2801 // SSE2 - Packed Mask Creation
2802 //===---------------------------------------------------------------------===//
2804 let ExeDomain = SSEPackedInt in {
2806 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2807 "pmovmskb\t{$src, $dst|$dst, $src}",
2808 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2809 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2810 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2811 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2812 "pmovmskb\t{$src, $dst|$dst, $src}",
2813 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2815 } // ExeDomain = SSEPackedInt
2817 //===---------------------------------------------------------------------===//
2818 // SSE2 - Conditional Store
2819 //===---------------------------------------------------------------------===//
2821 let ExeDomain = SSEPackedInt in {
2824 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2825 (ins VR128:$src, VR128:$mask),
2826 "maskmovdqu\t{$mask, $src|$src, $mask}",
2827 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2829 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2830 (ins VR128:$src, VR128:$mask),
2831 "maskmovdqu\t{$mask, $src|$src, $mask}",
2832 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2835 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2836 "maskmovdqu\t{$mask, $src|$src, $mask}",
2837 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2839 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2840 "maskmovdqu\t{$mask, $src|$src, $mask}",
2841 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2843 } // ExeDomain = SSEPackedInt
2845 //===---------------------------------------------------------------------===//
2846 // SSE2 - Move Doubleword
2847 //===---------------------------------------------------------------------===//
2849 // Move Int Doubleword to Packed Double Int
2850 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2851 "movd\t{$src, $dst|$dst, $src}",
2853 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2854 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2855 "movd\t{$src, $dst|$dst, $src}",
2857 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2859 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2860 "movd\t{$src, $dst|$dst, $src}",
2862 (v4i32 (scalar_to_vector GR32:$src)))]>;
2863 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2864 "movd\t{$src, $dst|$dst, $src}",
2866 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2867 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2868 "mov{d|q}\t{$src, $dst|$dst, $src}",
2870 (v2i64 (scalar_to_vector GR64:$src)))]>;
2871 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2872 "mov{d|q}\t{$src, $dst|$dst, $src}",
2873 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2876 // Move Int Doubleword to Single Scalar
2877 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2878 "movd\t{$src, $dst|$dst, $src}",
2879 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2881 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2882 "movd\t{$src, $dst|$dst, $src}",
2883 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2885 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2886 "movd\t{$src, $dst|$dst, $src}",
2887 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2889 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2890 "movd\t{$src, $dst|$dst, $src}",
2891 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2893 // Move Packed Doubleword Int to Packed Double Int
2894 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2895 "movd\t{$src, $dst|$dst, $src}",
2896 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2898 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2899 (ins i32mem:$dst, VR128:$src),
2900 "movd\t{$src, $dst|$dst, $src}",
2901 [(store (i32 (vector_extract (v4i32 VR128:$src),
2902 (iPTR 0))), addr:$dst)]>, VEX;
2903 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2904 "movd\t{$src, $dst|$dst, $src}",
2905 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2907 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2908 "movd\t{$src, $dst|$dst, $src}",
2909 [(store (i32 (vector_extract (v4i32 VR128:$src),
2910 (iPTR 0))), addr:$dst)]>;
2912 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2913 "mov{d|q}\t{$src, $dst|$dst, $src}",
2914 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2916 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2917 "movq\t{$src, $dst|$dst, $src}",
2918 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2920 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2921 "mov{d|q}\t{$src, $dst|$dst, $src}",
2922 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2923 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2924 "movq\t{$src, $dst|$dst, $src}",
2925 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2927 // Move Scalar Single to Double Int
2928 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2929 "movd\t{$src, $dst|$dst, $src}",
2930 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
2931 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2932 "movd\t{$src, $dst|$dst, $src}",
2933 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
2934 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2935 "movd\t{$src, $dst|$dst, $src}",
2936 [(set GR32:$dst, (bitconvert FR32:$src))]>;
2937 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
2938 "movd\t{$src, $dst|$dst, $src}",
2939 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
2941 // movd / movq to XMM register zero-extends
2942 let AddedComplexity = 15 in {
2943 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2944 "movd\t{$src, $dst|$dst, $src}",
2945 [(set VR128:$dst, (v4i32 (X86vzmovl
2946 (v4i32 (scalar_to_vector GR32:$src)))))]>,
2948 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2949 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2950 [(set VR128:$dst, (v2i64 (X86vzmovl
2951 (v2i64 (scalar_to_vector GR64:$src)))))]>,
2954 let AddedComplexity = 15 in {
2955 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2956 "movd\t{$src, $dst|$dst, $src}",
2957 [(set VR128:$dst, (v4i32 (X86vzmovl
2958 (v4i32 (scalar_to_vector GR32:$src)))))]>;
2959 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2960 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
2961 [(set VR128:$dst, (v2i64 (X86vzmovl
2962 (v2i64 (scalar_to_vector GR64:$src)))))]>;
2965 let AddedComplexity = 20 in {
2966 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2967 "movd\t{$src, $dst|$dst, $src}",
2969 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2970 (loadi32 addr:$src))))))]>,
2972 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2973 "movd\t{$src, $dst|$dst, $src}",
2975 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
2976 (loadi32 addr:$src))))))]>;
2978 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
2979 (MOVZDI2PDIrm addr:$src)>;
2980 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
2981 (MOVZDI2PDIrm addr:$src)>;
2982 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
2983 (MOVZDI2PDIrm addr:$src)>;
2986 // These are the correct encodings of the instructions so that we know how to
2987 // read correct assembly, even though we continue to emit the wrong ones for
2988 // compatibility with Darwin's buggy assembler.
2989 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2990 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
2991 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2992 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
2993 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2994 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
2995 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2996 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
2997 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
2998 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
2999 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3000 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3002 //===---------------------------------------------------------------------===//
3003 // SSE2 - Move Quadword
3004 //===---------------------------------------------------------------------===//
3006 // Move Quadword Int to Packed Quadword Int
3007 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3008 "vmovq\t{$src, $dst|$dst, $src}",
3010 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3011 VEX, Requires<[HasAVX]>;
3012 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3013 "movq\t{$src, $dst|$dst, $src}",
3015 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3016 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3018 // Move Packed Quadword Int to Quadword Int
3019 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3020 "movq\t{$src, $dst|$dst, $src}",
3021 [(store (i64 (vector_extract (v2i64 VR128:$src),
3022 (iPTR 0))), addr:$dst)]>, VEX;
3023 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3024 "movq\t{$src, $dst|$dst, $src}",
3025 [(store (i64 (vector_extract (v2i64 VR128:$src),
3026 (iPTR 0))), addr:$dst)]>;
3028 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3029 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3031 // Store / copy lower 64-bits of a XMM register.
3032 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3033 "movq\t{$src, $dst|$dst, $src}",
3034 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3035 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3036 "movq\t{$src, $dst|$dst, $src}",
3037 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3039 let AddedComplexity = 20 in
3040 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3041 "vmovq\t{$src, $dst|$dst, $src}",
3043 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3044 (loadi64 addr:$src))))))]>,
3045 XS, VEX, Requires<[HasAVX]>;
3047 let AddedComplexity = 20 in {
3048 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3049 "movq\t{$src, $dst|$dst, $src}",
3051 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3052 (loadi64 addr:$src))))))]>,
3053 XS, Requires<[HasSSE2]>;
3055 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3056 (MOVZQI2PQIrm addr:$src)>;
3057 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3058 (MOVZQI2PQIrm addr:$src)>;
3059 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3062 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3063 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3064 let AddedComplexity = 15 in
3065 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3066 "vmovq\t{$src, $dst|$dst, $src}",
3067 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3068 XS, VEX, Requires<[HasAVX]>;
3069 let AddedComplexity = 15 in
3070 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3071 "movq\t{$src, $dst|$dst, $src}",
3072 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3073 XS, Requires<[HasSSE2]>;
3075 let AddedComplexity = 20 in
3076 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3077 "vmovq\t{$src, $dst|$dst, $src}",
3078 [(set VR128:$dst, (v2i64 (X86vzmovl
3079 (loadv2i64 addr:$src))))]>,
3080 XS, VEX, Requires<[HasAVX]>;
3081 let AddedComplexity = 20 in {
3082 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3083 "movq\t{$src, $dst|$dst, $src}",
3084 [(set VR128:$dst, (v2i64 (X86vzmovl
3085 (loadv2i64 addr:$src))))]>,
3086 XS, Requires<[HasSSE2]>;
3088 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3089 (MOVZPQILo2PQIrm addr:$src)>;
3092 // Instructions to match in the assembler
3093 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3094 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3095 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3096 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3097 // Recognize "movd" with GR64 destination, but encode as a "movq"
3098 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3099 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3101 // Instructions for the disassembler
3102 // xr = XMM register
3105 let Predicates = [HasAVX] in
3106 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3107 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3108 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3109 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3111 //===---------------------------------------------------------------------===//
3112 // SSE2 - Misc Instructions
3113 //===---------------------------------------------------------------------===//
3116 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3117 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3118 TB, Requires<[HasSSE2]>;
3120 // Load, store, and memory fence
3121 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3122 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3123 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3124 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3125 def : Pat<(X86LFence), (LFENCE)>;
3126 def : Pat<(X86MFence), (MFENCE)>;
3129 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3130 // was introduced with SSE2, it's backward compatible.
3131 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3133 // Alias instructions that map zero vector to pxor / xorp* for sse.
3134 // We set canFoldAsLoad because this can be converted to a constant-pool
3135 // load of an all-ones value if folding it would be beneficial.
3136 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3137 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3138 // FIXME: Change encoding to pseudo.
3139 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3140 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3142 //===---------------------------------------------------------------------===//
3143 // SSE3 - Conversion Instructions
3144 //===---------------------------------------------------------------------===//
3146 // Convert Packed Double FP to Packed DW Integers
3147 let Predicates = [HasAVX] in {
3148 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3149 // register, but the same isn't true when using memory operands instead.
3150 // Provide other assembly rr and rm forms to address this explicitly.
3151 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3152 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3153 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3154 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3157 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3158 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3159 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3160 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3163 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3164 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3165 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3166 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3169 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3170 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3171 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3172 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3174 // Convert Packed DW Integers to Packed Double FP
3175 let Predicates = [HasAVX] in {
3176 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3177 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3178 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3179 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3180 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3181 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3182 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3183 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3186 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3187 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3188 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3189 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3191 // AVX 256-bit register conversion intrinsics
3192 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3193 (VCVTDQ2PDYrr VR128:$src)>;
3194 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3195 (VCVTDQ2PDYrm addr:$src)>;
3197 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3198 (VCVTPD2DQYrr VR256:$src)>;
3199 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3200 (VCVTPD2DQYrm addr:$src)>;
3202 //===---------------------------------------------------------------------===//
3203 // SSE3 - Move Instructions
3204 //===---------------------------------------------------------------------===//
3206 // Replicate Single FP
3207 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3208 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3209 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3210 [(set VR128:$dst, (v4f32 (rep_frag
3211 VR128:$src, (undef))))]>;
3212 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3213 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3214 [(set VR128:$dst, (rep_frag
3215 (memopv4f32 addr:$src), (undef)))]>;
3218 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3220 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3221 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3222 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3223 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3226 let Predicates = [HasAVX] in {
3227 // FIXME: Merge above classes when we have patterns for the ymm version
3228 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3229 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3230 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3231 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3233 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3234 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3236 // Replicate Double FP
3237 multiclass sse3_replicate_dfp<string OpcodeStr> {
3238 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3239 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3240 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3241 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3242 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3244 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3248 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3249 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3250 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3252 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3253 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3257 let Predicates = [HasAVX] in {
3258 // FIXME: Merge above classes when we have patterns for the ymm version
3259 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3260 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3262 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3264 // Move Unaligned Integer
3265 let Predicates = [HasAVX] in {
3266 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3267 "vlddqu\t{$src, $dst|$dst, $src}",
3268 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3269 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3270 "vlddqu\t{$src, $dst|$dst, $src}",
3271 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3273 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3274 "lddqu\t{$src, $dst|$dst, $src}",
3275 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3277 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3279 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3281 // Several Move patterns
3282 let AddedComplexity = 5 in {
3283 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3284 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3285 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3286 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3287 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3288 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3289 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3290 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3293 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3294 let AddedComplexity = 15 in
3295 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3296 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3297 let AddedComplexity = 20 in
3298 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3299 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3301 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3302 let AddedComplexity = 15 in
3303 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3304 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3305 let AddedComplexity = 20 in
3306 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3307 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3309 //===---------------------------------------------------------------------===//
3310 // SSE3 - Arithmetic
3311 //===---------------------------------------------------------------------===//
3313 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3314 X86MemOperand x86memop, bit Is2Addr = 1> {
3315 def rr : I<0xD0, MRMSrcReg,
3316 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3318 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3320 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3321 def rm : I<0xD0, MRMSrcMem,
3322 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3324 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3325 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3326 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3329 let Predicates = [HasAVX],
3330 ExeDomain = SSEPackedDouble in {
3331 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3332 f128mem, 0>, TB, XD, VEX_4V;
3333 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3334 f128mem, 0>, TB, OpSize, VEX_4V;
3335 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3336 f256mem, 0>, TB, XD, VEX_4V;
3337 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3338 f256mem, 0>, TB, OpSize, VEX_4V;
3340 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3341 ExeDomain = SSEPackedDouble in {
3342 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3344 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3345 f128mem>, TB, OpSize;
3348 //===---------------------------------------------------------------------===//
3349 // SSE3 Instructions
3350 //===---------------------------------------------------------------------===//
3353 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3354 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3355 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3357 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3359 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3361 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3363 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3364 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3365 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3367 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3368 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3369 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3371 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3372 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3373 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3375 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3377 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3379 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3382 let Predicates = [HasAVX] in {
3383 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3384 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3385 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3386 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3387 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3388 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3389 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3390 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3391 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3392 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3393 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3394 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3395 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3396 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3397 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3398 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3401 let Constraints = "$src1 = $dst" in {
3402 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3403 int_x86_sse3_hadd_ps>;
3404 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3405 int_x86_sse3_hadd_pd>;
3406 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3407 int_x86_sse3_hsub_ps>;
3408 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3409 int_x86_sse3_hsub_pd>;
3412 //===---------------------------------------------------------------------===//
3413 // SSSE3 - Packed Absolute Instructions
3414 //===---------------------------------------------------------------------===//
3417 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3418 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3419 PatFrag mem_frag128, Intrinsic IntId128> {
3420 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3422 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3423 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3426 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3428 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3431 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3434 let Predicates = [HasAVX] in {
3435 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3436 int_x86_ssse3_pabs_b_128>, VEX;
3437 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3438 int_x86_ssse3_pabs_w_128>, VEX;
3439 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3440 int_x86_ssse3_pabs_d_128>, VEX;
3443 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3444 int_x86_ssse3_pabs_b_128>;
3445 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3446 int_x86_ssse3_pabs_w_128>;
3447 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3448 int_x86_ssse3_pabs_d_128>;
3450 //===---------------------------------------------------------------------===//
3451 // SSSE3 - Packed Binary Operator Instructions
3452 //===---------------------------------------------------------------------===//
3454 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3455 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3456 PatFrag mem_frag128, Intrinsic IntId128,
3458 let isCommutable = 1 in
3459 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3460 (ins VR128:$src1, VR128:$src2),
3462 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3463 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3464 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3466 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3467 (ins VR128:$src1, i128mem:$src2),
3469 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3470 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3472 (IntId128 VR128:$src1,
3473 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3476 let Predicates = [HasAVX] in {
3477 let isCommutable = 0 in {
3478 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3479 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3480 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3481 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3482 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3483 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3484 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3485 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3486 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3487 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3488 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3489 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3490 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3491 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3492 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3493 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3494 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3495 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3496 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3497 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3498 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3499 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3501 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3502 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3505 // None of these have i8 immediate fields.
3506 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3507 let isCommutable = 0 in {
3508 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3509 int_x86_ssse3_phadd_w_128>;
3510 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3511 int_x86_ssse3_phadd_d_128>;
3512 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3513 int_x86_ssse3_phadd_sw_128>;
3514 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3515 int_x86_ssse3_phsub_w_128>;
3516 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3517 int_x86_ssse3_phsub_d_128>;
3518 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3519 int_x86_ssse3_phsub_sw_128>;
3520 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3521 int_x86_ssse3_pmadd_ub_sw_128>;
3522 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3523 int_x86_ssse3_pshuf_b_128>;
3524 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3525 int_x86_ssse3_psign_b_128>;
3526 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3527 int_x86_ssse3_psign_w_128>;
3528 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3529 int_x86_ssse3_psign_d_128>;
3531 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3532 int_x86_ssse3_pmul_hr_sw_128>;
3535 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3536 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3537 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3538 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3540 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3541 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3542 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3543 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3544 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3545 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3547 //===---------------------------------------------------------------------===//
3548 // SSSE3 - Packed Align Instruction Patterns
3549 //===---------------------------------------------------------------------===//
3551 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3552 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3553 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3555 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3557 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3559 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3560 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3562 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3564 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3568 let Predicates = [HasAVX] in
3569 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3570 let Constraints = "$src1 = $dst" in
3571 defm PALIGN : ssse3_palign<"palignr">;
3573 let AddedComplexity = 5 in {
3574 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3575 (PALIGNR128rr VR128:$src2, VR128:$src1,
3576 (SHUFFLE_get_palign_imm VR128:$src3))>,
3577 Requires<[HasSSSE3]>;
3578 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3579 (PALIGNR128rr VR128:$src2, VR128:$src1,
3580 (SHUFFLE_get_palign_imm VR128:$src3))>,
3581 Requires<[HasSSSE3]>;
3582 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3583 (PALIGNR128rr VR128:$src2, VR128:$src1,
3584 (SHUFFLE_get_palign_imm VR128:$src3))>,
3585 Requires<[HasSSSE3]>;
3586 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3587 (PALIGNR128rr VR128:$src2, VR128:$src1,
3588 (SHUFFLE_get_palign_imm VR128:$src3))>,
3589 Requires<[HasSSSE3]>;
3592 //===---------------------------------------------------------------------===//
3593 // SSSE3 Misc Instructions
3594 //===---------------------------------------------------------------------===//
3596 // Thread synchronization
3597 let usesCustomInserter = 1 in {
3598 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3599 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3600 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3601 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3604 let Uses = [EAX, ECX, EDX] in
3605 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3606 Requires<[HasSSE3]>;
3607 let Uses = [ECX, EAX] in
3608 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3609 Requires<[HasSSE3]>;
3611 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3612 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3614 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3615 Requires<[In32BitMode]>;
3616 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3617 Requires<[In64BitMode]>;
3619 //===---------------------------------------------------------------------===//
3620 // Non-Instruction Patterns
3621 //===---------------------------------------------------------------------===//
3623 // extload f32 -> f64. This matches load+fextend because we have a hack in
3624 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3626 // Since these loads aren't folded into the fextend, we have to match it
3628 let Predicates = [HasSSE2] in
3629 def : Pat<(fextend (loadf32 addr:$src)),
3630 (CVTSS2SDrm addr:$src)>;
3632 // FIXME: According to the intel manual, DEST[127:64] <- SRC1[127:64], while
3633 // in the non-AVX version bits 127:64 aren't touched. Find a better way to
3634 // represent this instead of always zeroing SRC1. One possible solution is
3635 // to represent the instruction w/ something similar as the "$src1 = $dst"
3636 // constraint but without the tied operands.
3637 let Predicates = [HasAVX] in
3638 def : Pat<(fextend (loadf32 addr:$src)),
3639 (VCVTSS2SDrm (f32 (EXTRACT_SUBREG (AVX_SET0PS), sub_ss)),
3643 let Predicates = [HasXMMInt] in {
3644 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3645 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3646 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3647 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3648 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3649 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3650 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3651 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3652 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3653 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3654 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3655 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3656 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3657 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3658 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3659 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3660 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3661 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3662 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3663 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3664 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3665 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3666 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3667 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3668 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3669 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3670 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3671 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3672 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3673 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3676 let Predicates = [HasAVX] in {
3677 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3678 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
3679 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
3680 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
3683 // Move scalar to XMM zero-extended
3684 // movd to XMM register zero-extends
3685 let AddedComplexity = 15 in {
3686 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3687 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3688 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3689 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3690 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3691 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3692 (MOVSSrr (v4f32 (V_SET0PS)),
3693 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3694 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3695 (MOVSSrr (v4i32 (V_SET0PI)),
3696 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3699 // Splat v2f64 / v2i64
3700 let AddedComplexity = 10 in {
3701 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3702 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3703 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3704 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3705 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3706 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3707 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3708 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3711 // Special unary SHUFPSrri case.
3712 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3713 (SHUFPSrri VR128:$src1, VR128:$src1,
3714 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3715 let AddedComplexity = 5 in
3716 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3717 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3718 Requires<[HasSSE2]>;
3719 // Special unary SHUFPDrri case.
3720 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3721 (SHUFPDrri VR128:$src1, VR128:$src1,
3722 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3723 Requires<[HasSSE2]>;
3724 // Special unary SHUFPDrri case.
3725 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3726 (SHUFPDrri VR128:$src1, VR128:$src1,
3727 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3728 Requires<[HasSSE2]>;
3729 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3730 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3731 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3732 Requires<[HasSSE2]>;
3734 // Special binary v4i32 shuffle cases with SHUFPS.
3735 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3736 (SHUFPSrri VR128:$src1, VR128:$src2,
3737 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3738 Requires<[HasSSE2]>;
3739 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3740 (SHUFPSrmi VR128:$src1, addr:$src2,
3741 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3742 Requires<[HasSSE2]>;
3743 // Special binary v2i64 shuffle cases using SHUFPDrri.
3744 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3745 (SHUFPDrri VR128:$src1, VR128:$src2,
3746 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3747 Requires<[HasSSE2]>;
3749 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3750 let AddedComplexity = 15 in {
3751 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3752 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3753 Requires<[OptForSpeed, HasSSE2]>;
3754 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3755 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3756 Requires<[OptForSpeed, HasSSE2]>;
3758 let AddedComplexity = 10 in {
3759 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3760 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3761 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3762 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3763 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3764 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3765 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3766 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3769 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3770 let AddedComplexity = 15 in {
3771 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3772 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3773 Requires<[OptForSpeed, HasSSE2]>;
3774 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3775 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3776 Requires<[OptForSpeed, HasSSE2]>;
3778 let AddedComplexity = 10 in {
3779 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3780 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3781 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3782 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3783 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3784 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3785 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3786 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3789 let AddedComplexity = 20 in {
3790 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3791 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3792 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3794 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3795 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3796 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3798 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3799 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3800 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3801 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3802 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3805 let AddedComplexity = 20 in {
3806 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3807 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3808 (MOVLPSrm VR128:$src1, addr:$src2)>;
3809 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3810 (MOVLPDrm VR128:$src1, addr:$src2)>;
3811 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3812 (MOVLPSrm VR128:$src1, addr:$src2)>;
3813 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3814 (MOVLPDrm VR128:$src1, addr:$src2)>;
3817 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3818 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3819 (MOVLPSmr addr:$src1, VR128:$src2)>;
3820 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3821 (MOVLPDmr addr:$src1, VR128:$src2)>;
3822 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3824 (MOVLPSmr addr:$src1, VR128:$src2)>;
3825 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3826 (MOVLPDmr addr:$src1, VR128:$src2)>;
3828 let AddedComplexity = 15 in {
3829 // Setting the lowest element in the vector.
3830 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3831 (MOVSSrr (v4i32 VR128:$src1),
3832 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3833 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3834 (MOVSDrr (v2i64 VR128:$src1),
3835 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3837 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3838 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3839 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3840 Requires<[HasSSE2]>;
3841 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3842 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3843 Requires<[HasSSE2]>;
3846 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3847 // fall back to this for SSE1)
3848 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3849 (SHUFPSrri VR128:$src2, VR128:$src1,
3850 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3852 // Set lowest element and zero upper elements.
3853 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3854 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3856 // vector -> vector casts
3857 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3858 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3859 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3860 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3862 // Use movaps / movups for SSE integer load / store (one byte shorter).
3863 let Predicates = [HasSSE1] in {
3864 def : Pat<(alignedloadv4i32 addr:$src),
3865 (MOVAPSrm addr:$src)>;
3866 def : Pat<(loadv4i32 addr:$src),
3867 (MOVUPSrm addr:$src)>;
3868 def : Pat<(alignedloadv2i64 addr:$src),
3869 (MOVAPSrm addr:$src)>;
3870 def : Pat<(loadv2i64 addr:$src),
3871 (MOVUPSrm addr:$src)>;
3873 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3874 (MOVAPSmr addr:$dst, VR128:$src)>;
3875 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3876 (MOVAPSmr addr:$dst, VR128:$src)>;
3877 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3878 (MOVAPSmr addr:$dst, VR128:$src)>;
3879 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3880 (MOVAPSmr addr:$dst, VR128:$src)>;
3881 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3882 (MOVUPSmr addr:$dst, VR128:$src)>;
3883 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3884 (MOVUPSmr addr:$dst, VR128:$src)>;
3885 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3886 (MOVUPSmr addr:$dst, VR128:$src)>;
3887 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3888 (MOVUPSmr addr:$dst, VR128:$src)>;
3891 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3892 let Predicates = [HasAVX] in {
3893 def : Pat<(alignedloadv4i32 addr:$src),
3894 (VMOVAPSrm addr:$src)>;
3895 def : Pat<(loadv4i32 addr:$src),
3896 (VMOVUPSrm addr:$src)>;
3897 def : Pat<(alignedloadv2i64 addr:$src),
3898 (VMOVAPSrm addr:$src)>;
3899 def : Pat<(loadv2i64 addr:$src),
3900 (VMOVUPSrm addr:$src)>;
3902 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3903 (VMOVAPSmr addr:$dst, VR128:$src)>;
3904 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3905 (VMOVAPSmr addr:$dst, VR128:$src)>;
3906 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3907 (VMOVAPSmr addr:$dst, VR128:$src)>;
3908 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3909 (VMOVAPSmr addr:$dst, VR128:$src)>;
3910 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3911 (VMOVUPSmr addr:$dst, VR128:$src)>;
3912 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3913 (VMOVUPSmr addr:$dst, VR128:$src)>;
3914 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3915 (VMOVUPSmr addr:$dst, VR128:$src)>;
3916 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3917 (VMOVUPSmr addr:$dst, VR128:$src)>;
3920 //===----------------------------------------------------------------------===//
3921 // SSE4.1 - Packed Move with Sign/Zero Extend
3922 //===----------------------------------------------------------------------===//
3924 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3925 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3926 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3927 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3929 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3930 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3932 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3936 let Predicates = [HasAVX] in {
3937 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3939 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3941 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3943 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3945 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3947 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3951 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3952 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3953 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3954 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3955 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3956 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3958 // Common patterns involving scalar load.
3959 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
3960 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3961 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
3962 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
3964 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
3965 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3966 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
3967 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
3969 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
3970 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3971 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
3972 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
3974 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
3975 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3976 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
3977 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
3979 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
3980 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3981 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
3982 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
3984 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
3985 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3986 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
3987 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
3990 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3991 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3992 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3993 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3995 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3996 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3998 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4002 let Predicates = [HasAVX] in {
4003 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4005 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4007 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4009 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4013 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4014 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4015 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4016 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4018 // Common patterns involving scalar load
4019 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4020 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4021 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4022 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4024 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4025 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4026 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4027 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4030 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4031 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4032 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4033 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4035 // Expecting a i16 load any extended to i32 value.
4036 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4037 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4038 [(set VR128:$dst, (IntId (bitconvert
4039 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4043 let Predicates = [HasAVX] in {
4044 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4046 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4049 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4050 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4052 // Common patterns involving scalar load
4053 def : Pat<(int_x86_sse41_pmovsxbq
4054 (bitconvert (v4i32 (X86vzmovl
4055 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4056 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4058 def : Pat<(int_x86_sse41_pmovzxbq
4059 (bitconvert (v4i32 (X86vzmovl
4060 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4061 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4063 //===----------------------------------------------------------------------===//
4064 // SSE4.1 - Extract Instructions
4065 //===----------------------------------------------------------------------===//
4067 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4068 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4069 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4070 (ins VR128:$src1, i32i8imm:$src2),
4071 !strconcat(OpcodeStr,
4072 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4073 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4075 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4076 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4077 !strconcat(OpcodeStr,
4078 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4081 // There's an AssertZext in the way of writing the store pattern
4082 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4085 let Predicates = [HasAVX] in {
4086 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4087 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4088 (ins VR128:$src1, i32i8imm:$src2),
4089 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4092 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4095 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4096 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4097 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4098 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4099 !strconcat(OpcodeStr,
4100 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4103 // There's an AssertZext in the way of writing the store pattern
4104 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4107 let Predicates = [HasAVX] in
4108 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4110 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4113 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4114 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4115 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4116 (ins VR128:$src1, i32i8imm:$src2),
4117 !strconcat(OpcodeStr,
4118 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4120 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4121 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4122 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4123 !strconcat(OpcodeStr,
4124 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4125 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4126 addr:$dst)]>, OpSize;
4129 let Predicates = [HasAVX] in
4130 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4132 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4134 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4135 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4136 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4137 (ins VR128:$src1, i32i8imm:$src2),
4138 !strconcat(OpcodeStr,
4139 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4141 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4142 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4143 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4144 !strconcat(OpcodeStr,
4145 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4146 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4147 addr:$dst)]>, OpSize, REX_W;
4150 let Predicates = [HasAVX] in
4151 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4153 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4155 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4157 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4158 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4159 (ins VR128:$src1, i32i8imm:$src2),
4160 !strconcat(OpcodeStr,
4161 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4163 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4165 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4166 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4167 !strconcat(OpcodeStr,
4168 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4169 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4170 addr:$dst)]>, OpSize;
4173 let Predicates = [HasAVX] in {
4174 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4175 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4176 (ins VR128:$src1, i32i8imm:$src2),
4177 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4180 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4182 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4183 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4186 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4187 Requires<[HasSSE41]>;
4189 //===----------------------------------------------------------------------===//
4190 // SSE4.1 - Insert Instructions
4191 //===----------------------------------------------------------------------===//
4193 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4194 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4195 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4197 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4199 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4201 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4202 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4203 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4205 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4207 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4209 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4210 imm:$src3))]>, OpSize;
4213 let Predicates = [HasAVX] in
4214 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4215 let Constraints = "$src1 = $dst" in
4216 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4218 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4219 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4220 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4222 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4224 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4226 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4228 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4229 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4231 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4233 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4235 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4236 imm:$src3)))]>, OpSize;
4239 let Predicates = [HasAVX] in
4240 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4241 let Constraints = "$src1 = $dst" in
4242 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4244 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4245 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4246 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4248 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4250 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4252 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4254 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4255 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4257 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4259 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4261 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4262 imm:$src3)))]>, OpSize;
4265 let Predicates = [HasAVX] in
4266 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4267 let Constraints = "$src1 = $dst" in
4268 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4270 // insertps has a few different modes, there's the first two here below which
4271 // are optimized inserts that won't zero arbitrary elements in the destination
4272 // vector. The next one matches the intrinsic and could zero arbitrary elements
4273 // in the target vector.
4274 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4275 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4276 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4278 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4280 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4282 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4284 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4285 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4287 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4289 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4291 (X86insrtps VR128:$src1,
4292 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4293 imm:$src3))]>, OpSize;
4296 let Constraints = "$src1 = $dst" in
4297 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4298 let Predicates = [HasAVX] in
4299 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4301 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4302 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4304 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4305 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4306 Requires<[HasSSE41]>;
4308 //===----------------------------------------------------------------------===//
4309 // SSE4.1 - Round Instructions
4310 //===----------------------------------------------------------------------===//
4312 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4313 X86MemOperand x86memop, RegisterClass RC,
4314 PatFrag mem_frag32, PatFrag mem_frag64,
4315 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4316 // Intrinsic operation, reg.
4317 // Vector intrinsic operation, reg
4318 def PSr : SS4AIi8<opcps, MRMSrcReg,
4319 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4320 !strconcat(OpcodeStr,
4321 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4322 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4325 // Vector intrinsic operation, mem
4326 def PSm : Ii8<opcps, MRMSrcMem,
4327 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4328 !strconcat(OpcodeStr,
4329 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4331 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4333 Requires<[HasSSE41]>;
4335 // Vector intrinsic operation, reg
4336 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4337 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4338 !strconcat(OpcodeStr,
4339 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4340 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4343 // Vector intrinsic operation, mem
4344 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4345 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4346 !strconcat(OpcodeStr,
4347 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4349 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4353 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4354 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4355 // Intrinsic operation, reg.
4356 // Vector intrinsic operation, reg
4357 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4358 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4359 !strconcat(OpcodeStr,
4360 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4363 // Vector intrinsic operation, mem
4364 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4365 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4366 !strconcat(OpcodeStr,
4367 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4368 []>, TA, OpSize, Requires<[HasSSE41]>;
4370 // Vector intrinsic operation, reg
4371 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4372 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4373 !strconcat(OpcodeStr,
4374 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4377 // Vector intrinsic operation, mem
4378 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4379 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4380 !strconcat(OpcodeStr,
4381 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4385 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4388 Intrinsic F64Int, bit Is2Addr = 1> {
4389 // Intrinsic operation, reg.
4390 def SSr : SS4AIi8<opcss, MRMSrcReg,
4391 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4393 !strconcat(OpcodeStr,
4394 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4395 !strconcat(OpcodeStr,
4396 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4397 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4400 // Intrinsic operation, mem.
4401 def SSm : SS4AIi8<opcss, MRMSrcMem,
4402 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4404 !strconcat(OpcodeStr,
4405 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4406 !strconcat(OpcodeStr,
4407 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4409 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4412 // Intrinsic operation, reg.
4413 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4414 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4416 !strconcat(OpcodeStr,
4417 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4418 !strconcat(OpcodeStr,
4419 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4420 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4423 // Intrinsic operation, mem.
4424 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4425 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4427 !strconcat(OpcodeStr,
4428 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4429 !strconcat(OpcodeStr,
4430 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4432 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4436 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4438 // Intrinsic operation, reg.
4439 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4440 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4441 !strconcat(OpcodeStr,
4442 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4445 // Intrinsic operation, mem.
4446 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4447 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4448 !strconcat(OpcodeStr,
4449 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4452 // Intrinsic operation, reg.
4453 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4454 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4455 !strconcat(OpcodeStr,
4456 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4459 // Intrinsic operation, mem.
4460 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4461 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4462 !strconcat(OpcodeStr,
4463 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4467 // FP round - roundss, roundps, roundsd, roundpd
4468 let Predicates = [HasAVX] in {
4470 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4471 memopv4f32, memopv2f64,
4472 int_x86_sse41_round_ps,
4473 int_x86_sse41_round_pd>, VEX;
4474 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4475 memopv8f32, memopv4f64,
4476 int_x86_avx_round_ps_256,
4477 int_x86_avx_round_pd_256>, VEX;
4478 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4479 int_x86_sse41_round_ss,
4480 int_x86_sse41_round_sd, 0>, VEX_4V;
4482 // Instructions for the assembler
4483 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4485 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4487 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4490 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4491 memopv4f32, memopv2f64,
4492 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4493 let Constraints = "$src1 = $dst" in
4494 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4495 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4497 //===----------------------------------------------------------------------===//
4498 // SSE4.1 - Packed Bit Test
4499 //===----------------------------------------------------------------------===//
4501 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4502 // the intel intrinsic that corresponds to this.
4503 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4504 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4505 "vptest\t{$src2, $src1|$src1, $src2}",
4506 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4508 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4509 "vptest\t{$src2, $src1|$src1, $src2}",
4510 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4513 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4514 "vptest\t{$src2, $src1|$src1, $src2}",
4515 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4517 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4518 "vptest\t{$src2, $src1|$src1, $src2}",
4519 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4523 let Defs = [EFLAGS] in {
4524 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4525 "ptest \t{$src2, $src1|$src1, $src2}",
4526 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4528 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4529 "ptest \t{$src2, $src1|$src1, $src2}",
4530 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4534 // The bit test instructions below are AVX only
4535 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4536 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4537 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4538 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4539 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4540 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4541 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4542 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4546 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4547 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4548 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4549 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4550 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4553 //===----------------------------------------------------------------------===//
4554 // SSE4.1 - Misc Instructions
4555 //===----------------------------------------------------------------------===//
4557 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4558 "popcnt{w}\t{$src, $dst|$dst, $src}",
4559 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4560 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4561 "popcnt{w}\t{$src, $dst|$dst, $src}",
4562 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4564 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4565 "popcnt{l}\t{$src, $dst|$dst, $src}",
4566 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4567 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4568 "popcnt{l}\t{$src, $dst|$dst, $src}",
4569 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4571 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4572 "popcnt{q}\t{$src, $dst|$dst, $src}",
4573 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4574 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4575 "popcnt{q}\t{$src, $dst|$dst, $src}",
4576 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4580 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4581 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4582 Intrinsic IntId128> {
4583 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4585 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4586 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4587 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4589 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4592 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4595 let Predicates = [HasAVX] in
4596 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4597 int_x86_sse41_phminposuw>, VEX;
4598 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4599 int_x86_sse41_phminposuw>;
4601 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4602 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4603 Intrinsic IntId128, bit Is2Addr = 1> {
4604 let isCommutable = 1 in
4605 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4606 (ins VR128:$src1, VR128:$src2),
4608 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4609 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4610 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4611 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4612 (ins VR128:$src1, i128mem:$src2),
4614 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4615 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4617 (IntId128 VR128:$src1,
4618 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4621 let Predicates = [HasAVX] in {
4622 let isCommutable = 0 in
4623 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4625 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4627 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4629 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4631 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4633 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4635 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4637 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4639 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4641 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4643 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4647 let Constraints = "$src1 = $dst" in {
4648 let isCommutable = 0 in
4649 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4650 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4651 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4652 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4653 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4654 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4655 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4656 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4657 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4658 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4659 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4662 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4663 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4664 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4665 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4667 /// SS48I_binop_rm - Simple SSE41 binary operator.
4668 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4669 ValueType OpVT, bit Is2Addr = 1> {
4670 let isCommutable = 1 in
4671 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4672 (ins VR128:$src1, VR128:$src2),
4674 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4675 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4676 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4678 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4679 (ins VR128:$src1, i128mem:$src2),
4681 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4682 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4683 [(set VR128:$dst, (OpNode VR128:$src1,
4684 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4688 let Predicates = [HasAVX] in
4689 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4690 let Constraints = "$src1 = $dst" in
4691 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4693 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4694 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4695 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4696 X86MemOperand x86memop, bit Is2Addr = 1> {
4697 let isCommutable = 1 in
4698 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4699 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4701 !strconcat(OpcodeStr,
4702 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4703 !strconcat(OpcodeStr,
4704 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4705 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4707 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4708 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4710 !strconcat(OpcodeStr,
4711 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4712 !strconcat(OpcodeStr,
4713 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4716 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4720 let Predicates = [HasAVX] in {
4721 let isCommutable = 0 in {
4722 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4723 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4724 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4725 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4726 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4727 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4728 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4729 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4730 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4731 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4732 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4733 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4735 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4736 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4737 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4738 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4739 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4740 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4743 let Constraints = "$src1 = $dst" in {
4744 let isCommutable = 0 in {
4745 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4746 VR128, memopv16i8, i128mem>;
4747 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4748 VR128, memopv16i8, i128mem>;
4749 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4750 VR128, memopv16i8, i128mem>;
4751 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4752 VR128, memopv16i8, i128mem>;
4754 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4755 VR128, memopv16i8, i128mem>;
4756 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4757 VR128, memopv16i8, i128mem>;
4760 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4761 let Predicates = [HasAVX] in {
4762 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4763 RegisterClass RC, X86MemOperand x86memop,
4764 PatFrag mem_frag, Intrinsic IntId> {
4765 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4766 (ins RC:$src1, RC:$src2, RC:$src3),
4767 !strconcat(OpcodeStr,
4768 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4769 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4770 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4772 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4773 (ins RC:$src1, x86memop:$src2, RC:$src3),
4774 !strconcat(OpcodeStr,
4775 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4777 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4779 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4783 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4784 memopv16i8, int_x86_sse41_blendvpd>;
4785 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4786 memopv16i8, int_x86_sse41_blendvps>;
4787 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4788 memopv16i8, int_x86_sse41_pblendvb>;
4789 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4790 memopv32i8, int_x86_avx_blendv_pd_256>;
4791 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4792 memopv32i8, int_x86_avx_blendv_ps_256>;
4794 /// SS41I_ternary_int - SSE 4.1 ternary operator
4795 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4796 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4797 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4798 (ins VR128:$src1, VR128:$src2),
4799 !strconcat(OpcodeStr,
4800 "\t{$src2, $dst|$dst, $src2}"),
4801 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4804 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4805 (ins VR128:$src1, i128mem:$src2),
4806 !strconcat(OpcodeStr,
4807 "\t{$src2, $dst|$dst, $src2}"),
4810 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4814 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4815 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4816 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4818 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
4819 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
4821 let Predicates = [HasAVX] in
4822 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4823 "vmovntdqa\t{$src, $dst|$dst, $src}",
4824 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4826 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4827 "movntdqa\t{$src, $dst|$dst, $src}",
4828 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4831 //===----------------------------------------------------------------------===//
4832 // SSE4.2 - Compare Instructions
4833 //===----------------------------------------------------------------------===//
4835 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4836 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4837 Intrinsic IntId128, bit Is2Addr = 1> {
4838 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4839 (ins VR128:$src1, VR128:$src2),
4841 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4842 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4843 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4845 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4846 (ins VR128:$src1, i128mem:$src2),
4848 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4849 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4851 (IntId128 VR128:$src1,
4852 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4855 let Predicates = [HasAVX] in
4856 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4858 let Constraints = "$src1 = $dst" in
4859 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4861 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4862 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4863 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4864 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4866 //===----------------------------------------------------------------------===//
4867 // SSE4.2 - String/text Processing Instructions
4868 //===----------------------------------------------------------------------===//
4870 // Packed Compare Implicit Length Strings, Return Mask
4871 multiclass pseudo_pcmpistrm<string asm> {
4872 def REG : PseudoI<(outs VR128:$dst),
4873 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4874 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4876 def MEM : PseudoI<(outs VR128:$dst),
4877 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4878 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4879 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4882 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4883 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4884 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4887 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
4888 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4889 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4890 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4891 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4892 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4893 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4896 let Defs = [XMM0, EFLAGS] in {
4897 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4898 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4899 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4900 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4901 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4902 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4905 // Packed Compare Explicit Length Strings, Return Mask
4906 multiclass pseudo_pcmpestrm<string asm> {
4907 def REG : PseudoI<(outs VR128:$dst),
4908 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4909 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4910 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4911 def MEM : PseudoI<(outs VR128:$dst),
4912 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4913 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4914 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4917 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4918 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4919 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4922 let Predicates = [HasAVX],
4923 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4924 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4925 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4926 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4927 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4928 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4929 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4932 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4933 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4934 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4935 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4936 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4937 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4938 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4941 // Packed Compare Implicit Length Strings, Return Index
4942 let Defs = [ECX, EFLAGS] in {
4943 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4944 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4945 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4946 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4947 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4948 (implicit EFLAGS)]>, OpSize;
4949 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4950 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4951 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4952 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4953 (implicit EFLAGS)]>, OpSize;
4957 let Predicates = [HasAVX] in {
4958 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4960 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
4962 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
4964 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
4966 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
4968 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
4972 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
4973 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
4974 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
4975 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
4976 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
4977 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
4979 // Packed Compare Explicit Length Strings, Return Index
4980 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
4981 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
4982 def rr : SS42AI<0x61, MRMSrcReg, (outs),
4983 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4984 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4985 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
4986 (implicit EFLAGS)]>, OpSize;
4987 def rm : SS42AI<0x61, MRMSrcMem, (outs),
4988 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4989 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
4991 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
4992 (implicit EFLAGS)]>, OpSize;
4996 let Predicates = [HasAVX] in {
4997 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
4999 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5001 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5003 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5005 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5007 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5011 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5012 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5013 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5014 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5015 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5016 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5018 //===----------------------------------------------------------------------===//
5019 // SSE4.2 - CRC Instructions
5020 //===----------------------------------------------------------------------===//
5022 // No CRC instructions have AVX equivalents
5024 // crc intrinsic instruction
5025 // This set of instructions are only rm, the only difference is the size
5027 let Constraints = "$src1 = $dst" in {
5028 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5029 (ins GR32:$src1, i8mem:$src2),
5030 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5032 (int_x86_sse42_crc32_32_8 GR32:$src1,
5033 (load addr:$src2)))]>;
5034 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5035 (ins GR32:$src1, GR8:$src2),
5036 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5038 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5039 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5040 (ins GR32:$src1, i16mem:$src2),
5041 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5043 (int_x86_sse42_crc32_32_16 GR32:$src1,
5044 (load addr:$src2)))]>,
5046 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5047 (ins GR32:$src1, GR16:$src2),
5048 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5050 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5052 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5053 (ins GR32:$src1, i32mem:$src2),
5054 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5056 (int_x86_sse42_crc32_32_32 GR32:$src1,
5057 (load addr:$src2)))]>;
5058 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5059 (ins GR32:$src1, GR32:$src2),
5060 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5062 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5063 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5064 (ins GR64:$src1, i8mem:$src2),
5065 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5067 (int_x86_sse42_crc32_64_8 GR64:$src1,
5068 (load addr:$src2)))]>,
5070 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5071 (ins GR64:$src1, GR8:$src2),
5072 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5074 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5076 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5077 (ins GR64:$src1, i64mem:$src2),
5078 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5080 (int_x86_sse42_crc32_64_64 GR64:$src1,
5081 (load addr:$src2)))]>,
5083 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5084 (ins GR64:$src1, GR64:$src2),
5085 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5087 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5091 //===----------------------------------------------------------------------===//
5092 // AES-NI Instructions
5093 //===----------------------------------------------------------------------===//
5095 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5096 Intrinsic IntId128, bit Is2Addr = 1> {
5097 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5098 (ins VR128:$src1, VR128:$src2),
5100 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5101 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5102 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5104 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5105 (ins VR128:$src1, i128mem:$src2),
5107 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5108 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5110 (IntId128 VR128:$src1,
5111 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5114 // Perform One Round of an AES Encryption/Decryption Flow
5115 let Predicates = [HasAVX, HasAES] in {
5116 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5117 int_x86_aesni_aesenc, 0>, VEX_4V;
5118 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5119 int_x86_aesni_aesenclast, 0>, VEX_4V;
5120 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5121 int_x86_aesni_aesdec, 0>, VEX_4V;
5122 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5123 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5126 let Constraints = "$src1 = $dst" in {
5127 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5128 int_x86_aesni_aesenc>;
5129 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5130 int_x86_aesni_aesenclast>;
5131 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5132 int_x86_aesni_aesdec>;
5133 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5134 int_x86_aesni_aesdeclast>;
5137 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5138 (AESENCrr VR128:$src1, VR128:$src2)>;
5139 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5140 (AESENCrm VR128:$src1, addr:$src2)>;
5141 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5142 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5143 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5144 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5145 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5146 (AESDECrr VR128:$src1, VR128:$src2)>;
5147 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5148 (AESDECrm VR128:$src1, addr:$src2)>;
5149 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5150 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5151 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5152 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5154 // Perform the AES InvMixColumn Transformation
5155 let Predicates = [HasAVX, HasAES] in {
5156 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5158 "vaesimc\t{$src1, $dst|$dst, $src1}",
5160 (int_x86_aesni_aesimc VR128:$src1))]>,
5162 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5163 (ins i128mem:$src1),
5164 "vaesimc\t{$src1, $dst|$dst, $src1}",
5166 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5169 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5171 "aesimc\t{$src1, $dst|$dst, $src1}",
5173 (int_x86_aesni_aesimc VR128:$src1))]>,
5175 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5176 (ins i128mem:$src1),
5177 "aesimc\t{$src1, $dst|$dst, $src1}",
5179 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5182 // AES Round Key Generation Assist
5183 let Predicates = [HasAVX, HasAES] in {
5184 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5185 (ins VR128:$src1, i8imm:$src2),
5186 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5188 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5190 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5191 (ins i128mem:$src1, i8imm:$src2),
5192 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5194 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5198 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5199 (ins VR128:$src1, i8imm:$src2),
5200 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5202 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5204 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5205 (ins i128mem:$src1, i8imm:$src2),
5206 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5208 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5212 //===----------------------------------------------------------------------===//
5213 // CLMUL Instructions
5214 //===----------------------------------------------------------------------===//
5216 // Carry-less Multiplication instructions
5217 let Constraints = "$src1 = $dst" in {
5218 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5219 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5220 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5223 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5224 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5225 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5229 // AVX carry-less Multiplication instructions
5230 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5231 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5232 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5235 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5236 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5237 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5241 multiclass pclmul_alias<string asm, int immop> {
5242 def : InstAlias<!strconcat("pclmul", asm,
5243 "dq {$src, $dst|$dst, $src}"),
5244 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5246 def : InstAlias<!strconcat("pclmul", asm,
5247 "dq {$src, $dst|$dst, $src}"),
5248 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5250 def : InstAlias<!strconcat("vpclmul", asm,
5251 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5252 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5254 def : InstAlias<!strconcat("vpclmul", asm,
5255 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5256 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5258 defm : pclmul_alias<"hqhq", 0x11>;
5259 defm : pclmul_alias<"hqlq", 0x01>;
5260 defm : pclmul_alias<"lqhq", 0x10>;
5261 defm : pclmul_alias<"lqlq", 0x00>;
5263 //===----------------------------------------------------------------------===//
5265 //===----------------------------------------------------------------------===//
5268 // Load from memory and broadcast to all elements of the destination operand
5269 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5270 X86MemOperand x86memop, Intrinsic Int> :
5271 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5272 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5273 [(set RC:$dst, (Int addr:$src))]>, VEX;
5275 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5276 int_x86_avx_vbroadcastss>;
5277 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5278 int_x86_avx_vbroadcastss_256>;
5279 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5280 int_x86_avx_vbroadcast_sd_256>;
5281 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5282 int_x86_avx_vbroadcastf128_pd_256>;
5284 // Insert packed floating-point values
5285 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5286 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5287 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5289 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5290 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5291 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5294 // Extract packed floating-point values
5295 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5296 (ins VR256:$src1, i8imm:$src2),
5297 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5299 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5300 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5301 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5304 // Conditional SIMD Packed Loads and Stores
5305 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5306 Intrinsic IntLd, Intrinsic IntLd256,
5307 Intrinsic IntSt, Intrinsic IntSt256,
5308 PatFrag pf128, PatFrag pf256> {
5309 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5310 (ins VR128:$src1, f128mem:$src2),
5311 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5312 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5314 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5315 (ins VR256:$src1, f256mem:$src2),
5316 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5317 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5319 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5320 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5322 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5323 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5324 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5325 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5326 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5329 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5330 int_x86_avx_maskload_ps,
5331 int_x86_avx_maskload_ps_256,
5332 int_x86_avx_maskstore_ps,
5333 int_x86_avx_maskstore_ps_256,
5334 memopv4f32, memopv8f32>;
5335 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5336 int_x86_avx_maskload_pd,
5337 int_x86_avx_maskload_pd_256,
5338 int_x86_avx_maskstore_pd,
5339 int_x86_avx_maskstore_pd_256,
5340 memopv2f64, memopv4f64>;
5342 // Permute Floating-Point Values
5343 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5344 RegisterClass RC, X86MemOperand x86memop_f,
5345 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5346 Intrinsic IntVar, Intrinsic IntImm> {
5347 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5348 (ins RC:$src1, RC:$src2),
5349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5350 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5351 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5352 (ins RC:$src1, x86memop_i:$src2),
5353 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5354 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5356 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5357 (ins RC:$src1, i8imm:$src2),
5358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5359 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5360 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5361 (ins x86memop_f:$src1, i8imm:$src2),
5362 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5363 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5366 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5367 memopv4f32, memopv4i32,
5368 int_x86_avx_vpermilvar_ps,
5369 int_x86_avx_vpermil_ps>;
5370 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5371 memopv8f32, memopv8i32,
5372 int_x86_avx_vpermilvar_ps_256,
5373 int_x86_avx_vpermil_ps_256>;
5374 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5375 memopv2f64, memopv2i64,
5376 int_x86_avx_vpermilvar_pd,
5377 int_x86_avx_vpermil_pd>;
5378 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5379 memopv4f64, memopv4i64,
5380 int_x86_avx_vpermilvar_pd_256,
5381 int_x86_avx_vpermil_pd_256>;
5383 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5384 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5385 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5387 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5388 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5389 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5392 // Zero All YMM registers
5393 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5394 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5396 // Zero Upper bits of YMM registers
5397 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5398 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5400 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5401 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5402 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5403 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5404 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5405 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5407 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5409 (VINSERTF128rr VR256:$src1, VR128:$src2,
5410 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5411 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5413 (VINSERTF128rr VR256:$src1, VR128:$src2,
5414 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5415 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5417 (VINSERTF128rr VR256:$src1, VR128:$src2,
5418 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5419 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5421 (VINSERTF128rr VR256:$src1, VR128:$src2,
5422 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5424 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5425 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5426 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5427 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5428 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5429 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5431 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5432 (v4f32 (VEXTRACTF128rr
5433 (v8f32 VR256:$src1),
5434 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5435 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5436 (v2f64 (VEXTRACTF128rr
5437 (v4f64 VR256:$src1),
5438 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5439 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5440 (v4i32 (VEXTRACTF128rr
5441 (v8i32 VR256:$src1),
5442 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5443 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5444 (v2i64 (VEXTRACTF128rr
5445 (v4i64 VR256:$src1),
5446 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5448 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5449 (VBROADCASTF128 addr:$src)>;
5451 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5452 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5453 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5454 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5455 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5456 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5458 def : Pat<(int_x86_avx_vperm2f128_ps_256
5459 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5460 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5461 def : Pat<(int_x86_avx_vperm2f128_pd_256
5462 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5463 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5464 def : Pat<(int_x86_avx_vperm2f128_si_256
5465 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5466 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5468 //===----------------------------------------------------------------------===//
5469 // SSE Shuffle pattern fragments
5470 //===----------------------------------------------------------------------===//
5472 // This is part of a "work in progress" refactoring. The idea is that all
5473 // vector shuffles are going to be translated into target specific nodes and
5474 // directly matched by the patterns below (which can be changed along the way)
5475 // The AVX version of some but not all of them are described here, and more
5476 // should come in a near future.
5478 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5479 // SSE2 loads, which are always promoted to v2i64. The last one should match
5480 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5481 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5482 // we investigate further.
5483 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5485 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5486 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5488 (PSHUFDmi addr:$src1, imm:$imm)>;
5489 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5491 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5493 // Shuffle with PSHUFD instruction.
5494 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5495 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5496 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5497 (PSHUFDri VR128:$src1, imm:$imm)>;
5499 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5500 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5501 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5502 (PSHUFDri VR128:$src1, imm:$imm)>;
5504 // Shuffle with SHUFPD instruction.
5505 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5506 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5507 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5508 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5509 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5510 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5512 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5513 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5514 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5515 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5517 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5518 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5519 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5520 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5522 // Shuffle with SHUFPS instruction.
5523 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5524 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5525 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5526 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5527 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5528 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5530 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5531 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5532 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5533 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5535 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5536 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5537 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5538 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5539 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5540 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5542 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5543 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5544 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5545 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5547 // Shuffle with MOVHLPS instruction
5548 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5549 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5550 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5551 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5553 // Shuffle with MOVDDUP instruction
5554 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5555 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5556 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5557 (MOVDDUPrm addr:$src)>;
5559 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5560 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5561 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5562 (MOVDDUPrm addr:$src)>;
5564 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5565 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5566 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5567 (MOVDDUPrm addr:$src)>;
5569 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5570 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5571 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5572 (MOVDDUPrm addr:$src)>;
5574 def : Pat<(X86Movddup (bc_v2f64
5575 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5576 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5577 def : Pat<(X86Movddup (bc_v2f64
5578 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5579 (MOVDDUPrm addr:$src)>;
5582 // Shuffle with UNPCKLPS
5583 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5584 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5585 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5586 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5587 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5588 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5590 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5591 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5592 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5593 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5594 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5595 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5597 // Shuffle with UNPCKHPS
5598 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5599 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5600 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5601 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5603 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5604 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5605 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5606 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5608 // Shuffle with UNPCKLPD
5609 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5610 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5611 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5612 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5613 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5614 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5616 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5617 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5618 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5619 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5620 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5621 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5623 // Shuffle with UNPCKHPD
5624 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5625 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5626 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5627 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5629 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5630 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5631 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5632 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5634 // Shuffle with PUNPCKLBW
5635 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5636 (bc_v16i8 (memopv2i64 addr:$src2)))),
5637 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5638 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5639 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5641 // Shuffle with PUNPCKLWD
5642 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5643 (bc_v8i16 (memopv2i64 addr:$src2)))),
5644 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5645 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5646 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5648 // Shuffle with PUNPCKLDQ
5649 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5650 (bc_v4i32 (memopv2i64 addr:$src2)))),
5651 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5652 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5653 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5655 // Shuffle with PUNPCKLQDQ
5656 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5657 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5658 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5659 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5661 // Shuffle with PUNPCKHBW
5662 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5663 (bc_v16i8 (memopv2i64 addr:$src2)))),
5664 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5665 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5666 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5668 // Shuffle with PUNPCKHWD
5669 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5670 (bc_v8i16 (memopv2i64 addr:$src2)))),
5671 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5672 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5673 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5675 // Shuffle with PUNPCKHDQ
5676 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5677 (bc_v4i32 (memopv2i64 addr:$src2)))),
5678 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5679 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5680 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5682 // Shuffle with PUNPCKHQDQ
5683 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5684 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5685 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5686 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5688 // Shuffle with MOVLHPS
5689 def : Pat<(X86Movlhps VR128:$src1,
5690 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5691 (MOVHPSrm VR128:$src1, addr:$src2)>;
5692 def : Pat<(X86Movlhps VR128:$src1,
5693 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5694 (MOVHPSrm VR128:$src1, addr:$src2)>;
5695 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5696 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5697 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5698 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5699 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5700 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5702 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd 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<(v2f64 (X86Movddup VR128:$src)),
5707 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5709 // Shuffle with MOVLHPD
5710 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5711 (scalar_to_vector (loadf64 addr:$src2)))),
5712 (MOVHPDrm VR128:$src1, addr:$src2)>;
5714 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5715 // is during lowering, where it's not possible to recognize the load fold cause
5716 // it has two uses through a bitcast. One use disappears at isel time and the
5717 // fold opportunity reappears.
5718 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5719 (scalar_to_vector (loadf64 addr:$src2)))),
5720 (MOVHPDrm VR128:$src1, addr:$src2)>;
5722 // Shuffle with MOVSS
5723 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5724 (MOVSSrr VR128:$src1, FR32:$src2)>;
5725 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5726 (MOVSSrr (v4i32 VR128:$src1),
5727 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5728 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5729 (MOVSSrr (v4f32 VR128:$src1),
5730 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5731 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5732 // is during lowering, where it's not possible to recognize the load fold cause
5733 // it has two uses through a bitcast. One use disappears at isel time and the
5734 // fold opportunity reappears.
5735 def : Pat<(X86Movss VR128:$src1,
5736 (bc_v4i32 (v2i64 (load addr:$src2)))),
5737 (MOVLPSrm VR128:$src1, addr:$src2)>;
5739 // Shuffle with MOVSD
5740 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5741 (MOVSDrr VR128:$src1, FR64:$src2)>;
5742 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5743 (MOVSDrr (v2i64 VR128:$src1),
5744 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5745 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5746 (MOVSDrr (v2f64 VR128:$src1),
5747 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5748 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5749 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5750 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5751 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5753 // Shuffle with MOVSHDUP
5754 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5755 (MOVSHDUPrr VR128:$src)>;
5756 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5757 (MOVSHDUPrm addr:$src)>;
5759 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5760 (MOVSHDUPrr VR128:$src)>;
5761 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5762 (MOVSHDUPrm addr:$src)>;
5764 // Shuffle with MOVSLDUP
5765 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5766 (MOVSLDUPrr VR128:$src)>;
5767 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5768 (MOVSLDUPrm addr:$src)>;
5770 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5771 (MOVSLDUPrr VR128:$src)>;
5772 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5773 (MOVSLDUPrm addr:$src)>;
5775 // Shuffle with PSHUFHW
5776 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5777 (PSHUFHWri VR128:$src, imm:$imm)>;
5778 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5779 (PSHUFHWmi addr:$src, imm:$imm)>;
5781 // Shuffle with PSHUFLW
5782 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5783 (PSHUFLWri VR128:$src, imm:$imm)>;
5784 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5785 (PSHUFLWmi addr:$src, imm:$imm)>;
5787 // Shuffle with PALIGN
5788 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5789 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5790 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5791 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5792 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5793 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5794 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5795 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5797 // Shuffle with MOVLPS
5798 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5799 (MOVLPSrm VR128:$src1, addr:$src2)>;
5800 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5801 (MOVLPSrm VR128:$src1, addr:$src2)>;
5802 def : Pat<(X86Movlps VR128:$src1,
5803 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5804 (MOVLPSrm VR128:$src1, addr:$src2)>;
5805 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5806 // is during lowering, where it's not possible to recognize the load fold cause
5807 // it has two uses through a bitcast. One use disappears at isel time and the
5808 // fold opportunity reappears.
5809 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5810 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5812 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
5813 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5815 // Shuffle with MOVLPD
5816 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5817 (MOVLPDrm VR128:$src1, addr:$src2)>;
5818 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5819 (MOVLPDrm VR128:$src1, addr:$src2)>;
5820 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5821 (scalar_to_vector (loadf64 addr:$src2)))),
5822 (MOVLPDrm VR128:$src1, addr:$src2)>;
5824 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5825 def : Pat<(store (f64 (vector_extract
5826 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5827 (MOVHPSmr addr:$dst, VR128:$src)>;
5828 def : Pat<(store (f64 (vector_extract
5829 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5830 (MOVHPDmr addr:$dst, VR128:$src)>;
5832 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5833 (MOVLPSmr addr:$src1, VR128:$src2)>;
5834 def : Pat<(store (v4i32 (X86Movlps
5835 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5836 (MOVLPSmr addr:$src1, VR128:$src2)>;
5838 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5839 (MOVLPDmr addr:$src1, VR128:$src2)>;
5840 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5841 (MOVLPDmr addr:$src1, VR128:$src2)>;