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:$src, sub_sd)>;
188 let AddedComplexity = 20 in {
189 let Predicates = [HasSSE1] in {
190 // MOVSSrm zeros the high parts of the register; represent this
191 // with SUBREG_TO_REG.
192 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
193 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
194 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
195 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
196 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
197 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
199 let Predicates = [HasSSE2] in {
200 // MOVSDrm zeros the high parts of the register; represent this
201 // with SUBREG_TO_REG.
202 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
203 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
204 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
205 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
206 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
207 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
208 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
209 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
210 def : Pat<(v2f64 (X86vzload addr:$src)),
211 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
215 let AddedComplexity = 20, Predicates = [HasAVX] in {
216 // MOVSSrm zeros the high parts of the register; represent this
217 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
218 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
219 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
220 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
221 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
222 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
223 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
224 // MOVSDrm zeros the high parts of the register; represent this
225 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
226 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
227 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
228 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
229 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
230 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
231 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
232 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
233 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
234 def : Pat<(v2f64 (X86vzload addr:$src)),
235 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
236 // Represent the same patterns above but in the form they appear for
238 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
239 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
240 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
241 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
242 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
243 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
246 // Store scalar value to memory.
247 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
248 "movss\t{$src, $dst|$dst, $src}",
249 [(store FR32:$src, addr:$dst)]>;
250 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
251 "movsd\t{$src, $dst|$dst, $src}",
252 [(store FR64:$src, addr:$dst)]>;
254 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
255 "movss\t{$src, $dst|$dst, $src}",
256 [(store FR32:$src, addr:$dst)]>, XS, VEX;
257 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
258 "movsd\t{$src, $dst|$dst, $src}",
259 [(store FR64:$src, addr:$dst)]>, XD, VEX;
261 // Extract and store.
262 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
265 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
266 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
269 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
271 // Move Aligned/Unaligned floating point values
272 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
273 X86MemOperand x86memop, PatFrag ld_frag,
274 string asm, Domain d,
275 bit IsReMaterializable = 1> {
276 let neverHasSideEffects = 1 in
277 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
278 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
279 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
280 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
281 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
282 [(set RC:$dst, (ld_frag addr:$src))], d>;
285 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
286 "movaps", SSEPackedSingle>, VEX;
287 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
288 "movapd", SSEPackedDouble>, OpSize, VEX;
289 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
290 "movups", SSEPackedSingle>, VEX;
291 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
292 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
294 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
295 "movaps", SSEPackedSingle>, VEX;
296 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
297 "movapd", SSEPackedDouble>, OpSize, VEX;
298 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
299 "movups", SSEPackedSingle>, VEX;
300 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
301 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
302 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
303 "movaps", SSEPackedSingle>, TB;
304 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
305 "movapd", SSEPackedDouble>, TB, OpSize;
306 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
307 "movups", SSEPackedSingle>, TB;
308 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
309 "movupd", SSEPackedDouble, 0>, TB, OpSize;
311 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
312 "movaps\t{$src, $dst|$dst, $src}",
313 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
314 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
315 "movapd\t{$src, $dst|$dst, $src}",
316 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
317 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
318 "movups\t{$src, $dst|$dst, $src}",
319 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
320 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
321 "movupd\t{$src, $dst|$dst, $src}",
322 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
323 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
324 "movaps\t{$src, $dst|$dst, $src}",
325 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
326 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
327 "movapd\t{$src, $dst|$dst, $src}",
328 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
329 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
330 "movups\t{$src, $dst|$dst, $src}",
331 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
332 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
333 "movupd\t{$src, $dst|$dst, $src}",
334 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
336 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
337 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
338 (VMOVUPSYmr addr:$dst, VR256:$src)>;
340 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
341 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
342 (VMOVUPDYmr addr:$dst, VR256:$src)>;
344 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
345 "movaps\t{$src, $dst|$dst, $src}",
346 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
347 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
348 "movapd\t{$src, $dst|$dst, $src}",
349 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
350 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
351 "movups\t{$src, $dst|$dst, $src}",
352 [(store (v4f32 VR128:$src), addr:$dst)]>;
353 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
354 "movupd\t{$src, $dst|$dst, $src}",
355 [(store (v2f64 VR128:$src), addr:$dst)]>;
357 // Intrinsic forms of MOVUPS/D load and store
358 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
359 (ins f128mem:$dst, VR128:$src),
360 "movups\t{$src, $dst|$dst, $src}",
361 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
362 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
363 (ins f128mem:$dst, VR128:$src),
364 "movupd\t{$src, $dst|$dst, $src}",
365 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
367 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
368 "movups\t{$src, $dst|$dst, $src}",
369 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
370 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
371 "movupd\t{$src, $dst|$dst, $src}",
372 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
374 // Move Low/High packed floating point values
375 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
376 PatFrag mov_frag, string base_opc,
378 def PSrm : PI<opc, MRMSrcMem,
379 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
380 !strconcat(base_opc, "s", asm_opr),
383 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
384 SSEPackedSingle>, TB;
386 def PDrm : PI<opc, MRMSrcMem,
387 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
388 !strconcat(base_opc, "d", asm_opr),
389 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
390 (scalar_to_vector (loadf64 addr:$src2)))))],
391 SSEPackedDouble>, TB, OpSize;
394 let AddedComplexity = 20 in {
395 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
396 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
397 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
398 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
400 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
401 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
402 "\t{$src2, $dst|$dst, $src2}">;
403 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
404 "\t{$src2, $dst|$dst, $src2}">;
407 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
408 "movlps\t{$src, $dst|$dst, $src}",
409 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
410 (iPTR 0))), addr:$dst)]>, VEX;
411 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
412 "movlpd\t{$src, $dst|$dst, $src}",
413 [(store (f64 (vector_extract (v2f64 VR128:$src),
414 (iPTR 0))), addr:$dst)]>, VEX;
415 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
416 "movlps\t{$src, $dst|$dst, $src}",
417 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
418 (iPTR 0))), addr:$dst)]>;
419 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
420 "movlpd\t{$src, $dst|$dst, $src}",
421 [(store (f64 (vector_extract (v2f64 VR128:$src),
422 (iPTR 0))), addr:$dst)]>;
424 // v2f64 extract element 1 is always custom lowered to unpack high to low
425 // and extract element 0 so the non-store version isn't too horrible.
426 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
427 "movhps\t{$src, $dst|$dst, $src}",
428 [(store (f64 (vector_extract
429 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
430 (undef)), (iPTR 0))), addr:$dst)]>,
432 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
433 "movhpd\t{$src, $dst|$dst, $src}",
434 [(store (f64 (vector_extract
435 (v2f64 (unpckh VR128:$src, (undef))),
436 (iPTR 0))), addr:$dst)]>,
438 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
439 "movhps\t{$src, $dst|$dst, $src}",
440 [(store (f64 (vector_extract
441 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
442 (undef)), (iPTR 0))), addr:$dst)]>;
443 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
444 "movhpd\t{$src, $dst|$dst, $src}",
445 [(store (f64 (vector_extract
446 (v2f64 (unpckh VR128:$src, (undef))),
447 (iPTR 0))), addr:$dst)]>;
449 let AddedComplexity = 20 in {
450 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
451 (ins VR128:$src1, VR128:$src2),
452 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
454 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
456 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
457 (ins VR128:$src1, VR128:$src2),
458 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
460 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
463 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
464 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
465 (ins VR128:$src1, VR128:$src2),
466 "movlhps\t{$src2, $dst|$dst, $src2}",
468 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
469 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
470 (ins VR128:$src1, VR128:$src2),
471 "movhlps\t{$src2, $dst|$dst, $src2}",
473 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
476 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
477 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
478 let AddedComplexity = 20 in {
479 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
480 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
481 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
482 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
485 //===----------------------------------------------------------------------===//
486 // SSE 1 & 2 - Conversion Instructions
487 //===----------------------------------------------------------------------===//
489 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
490 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
492 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
493 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
494 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
495 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
498 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
499 X86MemOperand x86memop, string asm> {
500 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
502 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
506 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
507 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
508 string asm, Domain d> {
509 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
510 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
511 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
512 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
515 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
516 X86MemOperand x86memop, string asm> {
517 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
518 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
519 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
520 (ins DstRC:$src1, x86memop:$src),
521 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
524 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
525 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
526 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
527 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
529 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
530 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
531 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
532 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
535 // The assembler can recognize rr 64-bit instructions by seeing a rxx
536 // register, but the same isn't true when only using memory operands,
537 // provide other assembly "l" and "q" forms to address this explicitly
538 // where appropriate to do so.
539 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
541 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
543 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
545 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
547 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
550 let Predicates = [HasAVX] in {
551 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
552 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
553 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
554 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
555 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
556 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
557 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
558 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
560 def : Pat<(f32 (sint_to_fp GR32:$src)),
561 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
562 def : Pat<(f32 (sint_to_fp GR64:$src)),
563 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
564 def : Pat<(f64 (sint_to_fp GR32:$src)),
565 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
566 def : Pat<(f64 (sint_to_fp GR64:$src)),
567 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
570 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
571 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
572 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
573 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
574 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
575 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
576 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
577 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
578 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
579 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
580 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
581 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
582 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
583 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
584 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
585 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
587 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
588 // and/or XMM operand(s).
590 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
591 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
593 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
594 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
595 [(set DstRC:$dst, (Int SrcRC:$src))]>;
596 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
597 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
598 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
601 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
602 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
603 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
604 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
606 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
607 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
608 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
609 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
610 (ins DstRC:$src1, x86memop:$src2),
612 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
613 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
614 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
617 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
618 f128mem, load, "cvtsd2si">, XD, VEX;
619 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
620 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
623 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
624 // Get rid of this hack or rename the intrinsics, there are several
625 // intructions that only match with the intrinsic form, why create duplicates
626 // to let them be recognized by the assembler?
627 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
628 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
629 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
630 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
631 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
632 f128mem, load, "cvtsd2si{l}">, XD;
633 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
634 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
637 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
638 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
639 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
640 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
642 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
643 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
644 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
645 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
648 let Constraints = "$src1 = $dst" in {
649 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
650 int_x86_sse_cvtsi2ss, i32mem, loadi32,
652 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
653 int_x86_sse_cvtsi642ss, i64mem, loadi64,
654 "cvtsi2ss{q}">, XS, REX_W;
655 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
656 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
658 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
659 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
660 "cvtsi2sd">, XD, REX_W;
665 // Aliases for intrinsics
666 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
667 f32mem, load, "cvttss2si">, XS, VEX;
668 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
669 int_x86_sse_cvttss2si64, f32mem, load,
670 "cvttss2si">, XS, VEX, VEX_W;
671 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
672 f128mem, load, "cvttsd2si">, XD, VEX;
673 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
674 int_x86_sse2_cvttsd2si64, f128mem, load,
675 "cvttsd2si">, XD, VEX, VEX_W;
676 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
677 f32mem, load, "cvttss2si">, XS;
678 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
679 int_x86_sse_cvttss2si64, f32mem, load,
680 "cvttss2si{q}">, XS, REX_W;
681 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
682 f128mem, load, "cvttsd2si">, XD;
683 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
684 int_x86_sse2_cvttsd2si64, f128mem, load,
685 "cvttsd2si{q}">, XD, REX_W;
687 let Pattern = []<dag> in {
688 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
689 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
690 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
691 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
693 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
694 "cvtdq2ps\t{$src, $dst|$dst, $src}",
695 SSEPackedSingle>, TB, VEX;
696 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
697 "cvtdq2ps\t{$src, $dst|$dst, $src}",
698 SSEPackedSingle>, TB, VEX;
701 let Pattern = []<dag> in {
702 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
703 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
704 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
705 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
706 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
707 "cvtdq2ps\t{$src, $dst|$dst, $src}",
708 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
711 let Predicates = [HasSSE1] in {
712 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
713 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
714 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
715 (CVTSS2SIrm addr:$src)>;
716 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
717 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
718 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
719 (CVTSS2SI64rm addr:$src)>;
722 let Predicates = [HasAVX] in {
723 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
724 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
725 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
726 (VCVTSS2SIrm addr:$src)>;
727 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
728 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
729 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
730 (VCVTSS2SI64rm addr:$src)>;
735 // Convert scalar double to scalar single
736 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
737 (ins FR64:$src1, FR64:$src2),
738 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
740 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
741 (ins FR64:$src1, f64mem:$src2),
742 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
743 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
744 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
747 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
748 "cvtsd2ss\t{$src, $dst|$dst, $src}",
749 [(set FR32:$dst, (fround FR64:$src))]>;
750 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
751 "cvtsd2ss\t{$src, $dst|$dst, $src}",
752 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
753 Requires<[HasSSE2, OptForSize]>;
755 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
756 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
758 let Constraints = "$src1 = $dst" in
759 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
760 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
762 // Convert scalar single to scalar double
763 // SSE2 instructions with XS prefix
764 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
765 (ins FR32:$src1, FR32:$src2),
766 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
767 []>, XS, Requires<[HasAVX]>, VEX_4V;
768 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
769 (ins FR32:$src1, f32mem:$src2),
770 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
771 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
773 let Predicates = [HasAVX] in {
774 def : Pat<(f64 (fextend FR32:$src)),
775 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
776 def : Pat<(fextend (loadf32 addr:$src)),
777 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
778 def : Pat<(extloadf32 addr:$src),
779 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
782 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
783 "cvtss2sd\t{$src, $dst|$dst, $src}",
784 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
786 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
787 "cvtss2sd\t{$src, $dst|$dst, $src}",
788 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
789 Requires<[HasSSE2, OptForSize]>;
791 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
792 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
793 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
794 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
795 VR128:$src2))]>, XS, VEX_4V,
797 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
798 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
799 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
800 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
801 (load addr:$src2)))]>, XS, VEX_4V,
803 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
804 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
805 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
806 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
807 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
810 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
811 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
812 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
813 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
814 (load addr:$src2)))]>, XS,
818 def : Pat<(extloadf32 addr:$src),
819 (CVTSS2SDrr (MOVSSrm addr:$src))>,
820 Requires<[HasSSE2, OptForSpeed]>;
822 // Convert doubleword to packed single/double fp
823 // SSE2 instructions without OpSize prefix
824 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
825 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
826 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
827 TB, VEX, Requires<[HasAVX]>;
828 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
829 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
830 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
831 (bitconvert (memopv2i64 addr:$src))))]>,
832 TB, VEX, Requires<[HasAVX]>;
833 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
834 "cvtdq2ps\t{$src, $dst|$dst, $src}",
835 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
836 TB, Requires<[HasSSE2]>;
837 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
838 "cvtdq2ps\t{$src, $dst|$dst, $src}",
839 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
840 (bitconvert (memopv2i64 addr:$src))))]>,
841 TB, Requires<[HasSSE2]>;
843 // FIXME: why the non-intrinsic version is described as SSE3?
844 // SSE2 instructions with XS prefix
845 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
846 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
847 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
848 XS, VEX, Requires<[HasAVX]>;
849 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
850 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
851 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
852 (bitconvert (memopv2i64 addr:$src))))]>,
853 XS, VEX, Requires<[HasAVX]>;
854 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
855 "cvtdq2pd\t{$src, $dst|$dst, $src}",
856 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
857 XS, Requires<[HasSSE2]>;
858 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
859 "cvtdq2pd\t{$src, $dst|$dst, $src}",
860 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
861 (bitconvert (memopv2i64 addr:$src))))]>,
862 XS, Requires<[HasSSE2]>;
865 // Convert packed single/double fp to doubleword
866 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
867 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
868 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
869 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
870 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
871 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
872 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
873 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
874 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
875 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
876 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
877 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
879 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
880 "cvtps2dq\t{$src, $dst|$dst, $src}",
881 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
883 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
885 "cvtps2dq\t{$src, $dst|$dst, $src}",
886 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
887 (memop addr:$src)))]>, VEX;
888 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
889 "cvtps2dq\t{$src, $dst|$dst, $src}",
890 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
891 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
892 "cvtps2dq\t{$src, $dst|$dst, $src}",
893 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
894 (memop addr:$src)))]>;
896 // SSE2 packed instructions with XD prefix
897 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
898 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
899 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
900 XD, VEX, Requires<[HasAVX]>;
901 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
902 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
903 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
904 (memop addr:$src)))]>,
905 XD, VEX, Requires<[HasAVX]>;
906 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
907 "cvtpd2dq\t{$src, $dst|$dst, $src}",
908 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
909 XD, Requires<[HasSSE2]>;
910 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
911 "cvtpd2dq\t{$src, $dst|$dst, $src}",
912 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
913 (memop addr:$src)))]>,
914 XD, Requires<[HasSSE2]>;
917 // Convert with truncation packed single/double fp to doubleword
918 // SSE2 packed instructions with XS prefix
919 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
920 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
921 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
922 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
923 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
924 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
925 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
926 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
927 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
928 "cvttps2dq\t{$src, $dst|$dst, $src}",
930 (int_x86_sse2_cvttps2dq VR128:$src))]>;
931 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
932 "cvttps2dq\t{$src, $dst|$dst, $src}",
934 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
936 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
937 "vcvttps2dq\t{$src, $dst|$dst, $src}",
939 (int_x86_sse2_cvttps2dq VR128:$src))]>,
940 XS, VEX, Requires<[HasAVX]>;
941 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
942 "vcvttps2dq\t{$src, $dst|$dst, $src}",
943 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
944 (memop addr:$src)))]>,
945 XS, VEX, Requires<[HasAVX]>;
947 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
948 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
949 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
950 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
952 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
953 (Int_VCVTDQ2PSrr VR128:$src)>, Requires<[HasAVX]>;
954 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
955 (VCVTTPS2DQrr VR128:$src)>, Requires<[HasAVX]>;
956 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
957 (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
958 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
959 (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
961 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
963 "cvttpd2dq\t{$src, $dst|$dst, $src}",
964 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
966 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
968 "cvttpd2dq\t{$src, $dst|$dst, $src}",
969 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
970 (memop addr:$src)))]>, VEX;
971 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
972 "cvttpd2dq\t{$src, $dst|$dst, $src}",
973 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
974 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
975 "cvttpd2dq\t{$src, $dst|$dst, $src}",
976 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
977 (memop addr:$src)))]>;
979 // The assembler can recognize rr 256-bit instructions by seeing a ymm
980 // register, but the same isn't true when using memory operands instead.
981 // Provide other assembly rr and rm forms to address this explicitly.
982 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
983 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
984 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
985 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
988 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
989 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
990 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
991 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
994 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
995 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
996 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
997 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
999 // Convert packed single to packed double
1000 let Predicates = [HasAVX] in {
1001 // SSE2 instructions without OpSize prefix
1002 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1003 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1004 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1005 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1006 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1007 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1008 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1009 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1011 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1012 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1013 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1014 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1016 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1017 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1018 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1019 VEX, Requires<[HasAVX]>;
1020 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1021 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1022 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1023 (load addr:$src)))]>,
1024 VEX, Requires<[HasAVX]>;
1025 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1026 "cvtps2pd\t{$src, $dst|$dst, $src}",
1027 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1028 TB, Requires<[HasSSE2]>;
1029 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1030 "cvtps2pd\t{$src, $dst|$dst, $src}",
1031 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1032 (load addr:$src)))]>,
1033 TB, Requires<[HasSSE2]>;
1035 // Convert packed double to packed single
1036 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1037 // register, but the same isn't true when using memory operands instead.
1038 // Provide other assembly rr and rm forms to address this explicitly.
1039 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1040 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1041 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1042 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1045 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1046 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1047 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1048 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1051 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1052 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1053 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1054 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1055 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1056 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1057 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1058 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1061 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1062 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1063 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1064 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1066 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1067 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1068 (memop addr:$src)))]>;
1069 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1070 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1071 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1072 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1073 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1074 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1075 (memop addr:$src)))]>;
1077 // AVX 256-bit register conversion intrinsics
1078 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1079 // whenever possible to avoid declaring two versions of each one.
1080 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1081 (VCVTDQ2PSYrr VR256:$src)>;
1082 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1083 (VCVTDQ2PSYrm addr:$src)>;
1085 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1086 (VCVTPD2PSYrr VR256:$src)>;
1087 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1088 (VCVTPD2PSYrm addr:$src)>;
1090 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1091 (VCVTPS2DQYrr VR256:$src)>;
1092 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1093 (VCVTPS2DQYrm addr:$src)>;
1095 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1096 (VCVTPS2PDYrr VR128:$src)>;
1097 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1098 (VCVTPS2PDYrm addr:$src)>;
1100 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1101 (VCVTTPD2DQYrr VR256:$src)>;
1102 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1103 (VCVTTPD2DQYrm addr:$src)>;
1105 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1106 (VCVTTPS2DQYrr VR256:$src)>;
1107 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1108 (VCVTTPS2DQYrm addr:$src)>;
1110 // Match fround and fextend for 128/256-bit conversions
1111 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1112 (VCVTPD2PSYrr VR256:$src)>;
1113 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1114 (VCVTPD2PSYrm addr:$src)>;
1116 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1117 (VCVTPS2PDYrr VR128:$src)>;
1118 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1119 (VCVTPS2PDYrm addr:$src)>;
1121 //===----------------------------------------------------------------------===//
1122 // SSE 1 & 2 - Compare Instructions
1123 //===----------------------------------------------------------------------===//
1125 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1126 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1127 string asm, string asm_alt> {
1128 let isAsmParserOnly = 1 in {
1129 def rr : SIi8<0xC2, MRMSrcReg,
1130 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1133 def rm : SIi8<0xC2, MRMSrcMem,
1134 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1138 // Accept explicit immediate argument form instead of comparison code.
1139 def rr_alt : SIi8<0xC2, MRMSrcReg,
1140 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1143 def rm_alt : SIi8<0xC2, MRMSrcMem,
1144 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1148 let neverHasSideEffects = 1 in {
1149 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1150 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1151 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1153 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1154 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1155 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1159 let Constraints = "$src1 = $dst" in {
1160 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1161 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1162 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1163 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1164 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1165 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1166 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1167 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1168 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1169 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1170 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1171 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1172 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1173 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1174 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1175 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1177 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1178 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1179 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1180 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1181 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1182 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1183 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1184 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1185 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1186 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1187 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1188 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1189 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1192 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1193 Intrinsic Int, string asm> {
1194 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1195 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1196 [(set VR128:$dst, (Int VR128:$src1,
1197 VR128:$src, imm:$cc))]>;
1198 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1199 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1200 [(set VR128:$dst, (Int VR128:$src1,
1201 (load addr:$src), imm:$cc))]>;
1204 // Aliases to match intrinsics which expect XMM operand(s).
1205 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1206 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1208 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1209 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1211 let Constraints = "$src1 = $dst" in {
1212 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1213 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1214 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1215 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1219 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1220 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1221 ValueType vt, X86MemOperand x86memop,
1222 PatFrag ld_frag, string OpcodeStr, Domain d> {
1223 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1224 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1225 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1226 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1227 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1228 [(set EFLAGS, (OpNode (vt RC:$src1),
1229 (ld_frag addr:$src2)))], d>;
1232 let Defs = [EFLAGS] in {
1233 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1234 "ucomiss", SSEPackedSingle>, VEX;
1235 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1236 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1237 let Pattern = []<dag> in {
1238 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1239 "comiss", SSEPackedSingle>, VEX;
1240 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1241 "comisd", SSEPackedDouble>, OpSize, VEX;
1244 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1245 load, "ucomiss", SSEPackedSingle>, VEX;
1246 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1247 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1249 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1250 load, "comiss", SSEPackedSingle>, VEX;
1251 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1252 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1253 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1254 "ucomiss", SSEPackedSingle>, TB;
1255 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1256 "ucomisd", SSEPackedDouble>, TB, OpSize;
1258 let Pattern = []<dag> in {
1259 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1260 "comiss", SSEPackedSingle>, TB;
1261 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1262 "comisd", SSEPackedDouble>, TB, OpSize;
1265 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1266 load, "ucomiss", SSEPackedSingle>, TB;
1267 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1268 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1270 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1271 "comiss", SSEPackedSingle>, TB;
1272 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1273 "comisd", SSEPackedDouble>, TB, OpSize;
1274 } // Defs = [EFLAGS]
1276 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1277 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1278 Intrinsic Int, string asm, string asm_alt,
1280 let isAsmParserOnly = 1 in {
1281 def rri : PIi8<0xC2, MRMSrcReg,
1282 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1283 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1284 def rmi : PIi8<0xC2, MRMSrcMem,
1285 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1286 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1289 // Accept explicit immediate argument form instead of comparison code.
1290 def rri_alt : PIi8<0xC2, MRMSrcReg,
1291 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1293 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1294 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1298 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1299 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1300 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1301 SSEPackedSingle>, VEX_4V;
1302 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1303 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1304 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1305 SSEPackedDouble>, OpSize, VEX_4V;
1306 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1307 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1308 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1309 SSEPackedSingle>, VEX_4V;
1310 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1311 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1312 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1313 SSEPackedDouble>, OpSize, VEX_4V;
1314 let Constraints = "$src1 = $dst" in {
1315 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1316 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1317 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1318 SSEPackedSingle>, TB;
1319 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1320 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1321 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1322 SSEPackedDouble>, TB, OpSize;
1325 let Predicates = [HasSSE1] in {
1326 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1327 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1328 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1329 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1332 let Predicates = [HasSSE2] in {
1333 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1334 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1335 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1336 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1339 let Predicates = [HasAVX] in {
1340 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1341 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1342 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1343 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1344 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1345 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1346 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1347 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1349 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
1350 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
1351 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
1352 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
1353 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
1354 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
1355 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
1356 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
1359 //===----------------------------------------------------------------------===//
1360 // SSE 1 & 2 - Shuffle Instructions
1361 //===----------------------------------------------------------------------===//
1363 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1364 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1365 ValueType vt, string asm, PatFrag mem_frag,
1366 Domain d, bit IsConvertibleToThreeAddress = 0> {
1367 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1368 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1369 [(set RC:$dst, (vt (shufp:$src3
1370 RC:$src1, (mem_frag addr:$src2))))], d>;
1371 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1372 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1373 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1375 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1378 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1379 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1380 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1381 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1382 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1383 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1384 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1385 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1386 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1387 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1388 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1389 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1391 let Constraints = "$src1 = $dst" in {
1392 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1393 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1394 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1396 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1397 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1398 memopv2f64, SSEPackedDouble>, TB, OpSize;
1401 //===----------------------------------------------------------------------===//
1402 // SSE 1 & 2 - Unpack Instructions
1403 //===----------------------------------------------------------------------===//
1405 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1406 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1407 PatFrag mem_frag, RegisterClass RC,
1408 X86MemOperand x86memop, string asm,
1410 def rr : PI<opc, MRMSrcReg,
1411 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1413 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1414 def rm : PI<opc, MRMSrcMem,
1415 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1417 (vt (OpNode RC:$src1,
1418 (mem_frag addr:$src2))))], d>;
1421 let AddedComplexity = 10 in {
1422 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1423 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1424 SSEPackedSingle>, VEX_4V;
1425 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1426 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1427 SSEPackedDouble>, OpSize, VEX_4V;
1428 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1429 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1430 SSEPackedSingle>, VEX_4V;
1431 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1432 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1433 SSEPackedDouble>, OpSize, VEX_4V;
1435 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1436 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1437 SSEPackedSingle>, VEX_4V;
1438 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1439 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1440 SSEPackedDouble>, OpSize, VEX_4V;
1441 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1442 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1443 SSEPackedSingle>, VEX_4V;
1444 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1445 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1446 SSEPackedDouble>, OpSize, VEX_4V;
1448 let Constraints = "$src1 = $dst" in {
1449 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1450 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1451 SSEPackedSingle>, TB;
1452 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1453 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1454 SSEPackedDouble>, TB, OpSize;
1455 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1456 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1457 SSEPackedSingle>, TB;
1458 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1459 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1460 SSEPackedDouble>, TB, OpSize;
1461 } // Constraints = "$src1 = $dst"
1462 } // AddedComplexity
1464 //===----------------------------------------------------------------------===//
1465 // SSE 1 & 2 - Extract Floating-Point Sign mask
1466 //===----------------------------------------------------------------------===//
1468 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1469 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1471 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1472 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1473 [(set GR32:$dst, (Int RC:$src))], d>;
1474 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1475 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1478 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1479 SSEPackedSingle>, TB;
1480 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1481 SSEPackedDouble>, TB, OpSize;
1483 def : Pat<(i32 (X86fgetsign FR32:$src)),
1484 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1485 sub_ss))>, Requires<[HasSSE1]>;
1486 def : Pat<(i64 (X86fgetsign FR32:$src)),
1487 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1488 sub_ss))>, Requires<[HasSSE1]>;
1489 def : Pat<(i32 (X86fgetsign FR64:$src)),
1490 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1491 sub_sd))>, Requires<[HasSSE2]>;
1492 def : Pat<(i64 (X86fgetsign FR64:$src)),
1493 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1494 sub_sd))>, Requires<[HasSSE2]>;
1496 let Predicates = [HasAVX] in {
1497 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1498 "movmskps", SSEPackedSingle>, TB, VEX;
1499 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1500 "movmskpd", SSEPackedDouble>, TB, OpSize,
1502 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1503 "movmskps", SSEPackedSingle>, TB, VEX;
1504 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1505 "movmskpd", SSEPackedDouble>, TB, OpSize,
1508 def : Pat<(i32 (X86fgetsign FR32:$src)),
1509 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1511 def : Pat<(i64 (X86fgetsign FR32:$src)),
1512 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1514 def : Pat<(i32 (X86fgetsign FR64:$src)),
1515 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1517 def : Pat<(i64 (X86fgetsign FR64:$src)),
1518 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1522 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1523 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1524 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1525 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1527 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1528 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1529 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1530 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1534 //===----------------------------------------------------------------------===//
1535 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1536 //===----------------------------------------------------------------------===//
1538 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1539 // names that start with 'Fs'.
1541 // Alias instructions that map fld0 to pxor for sse.
1542 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1543 canFoldAsLoad = 1 in {
1544 // FIXME: Set encoding to pseudo!
1545 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1546 [(set FR32:$dst, fp32imm0)]>,
1547 Requires<[HasSSE1]>, TB, OpSize;
1548 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1549 [(set FR64:$dst, fpimm0)]>,
1550 Requires<[HasSSE2]>, TB, OpSize;
1551 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1552 [(set FR32:$dst, fp32imm0)]>,
1553 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1554 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1555 [(set FR64:$dst, fpimm0)]>,
1556 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1559 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1560 // bits are disregarded.
1561 let neverHasSideEffects = 1 in {
1562 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1563 "movaps\t{$src, $dst|$dst, $src}", []>;
1564 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1565 "movapd\t{$src, $dst|$dst, $src}", []>;
1568 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1569 // bits are disregarded.
1570 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1571 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1572 "movaps\t{$src, $dst|$dst, $src}",
1573 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1574 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1575 "movapd\t{$src, $dst|$dst, $src}",
1576 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1579 //===----------------------------------------------------------------------===//
1580 // SSE 1 & 2 - Logical Instructions
1581 //===----------------------------------------------------------------------===//
1583 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1585 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1587 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1588 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1590 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1591 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1593 let Constraints = "$src1 = $dst" in {
1594 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1595 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1597 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1598 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1602 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1603 let mayLoad = 0 in {
1604 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1605 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1606 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1609 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1610 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1612 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1614 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1616 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
1617 // are all promoted to v2i64, and the patterns are covered by the int
1618 // version. This is needed in SSE only, because v2i64 isn't supported on
1619 // SSE1, but only on SSE2.
1620 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1621 !strconcat(OpcodeStr, "ps"), f128mem, [],
1622 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1623 (memopv2i64 addr:$src2)))], 0>, VEX_4V;
1625 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1626 !strconcat(OpcodeStr, "pd"), f128mem,
1627 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1628 (bc_v2i64 (v2f64 VR128:$src2))))],
1629 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1630 (memopv2i64 addr:$src2)))], 0>,
1632 let Constraints = "$src1 = $dst" in {
1633 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1634 !strconcat(OpcodeStr, "ps"), f128mem,
1635 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1636 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1637 (memopv2i64 addr:$src2)))]>, TB;
1639 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1640 !strconcat(OpcodeStr, "pd"), f128mem,
1641 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1642 (bc_v2i64 (v2f64 VR128:$src2))))],
1643 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1644 (memopv2i64 addr:$src2)))]>, TB, OpSize;
1648 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1650 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1652 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1653 !strconcat(OpcodeStr, "ps"), f256mem,
1654 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
1655 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1656 (memopv4i64 addr:$src2)))], 0>, VEX_4V;
1658 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1659 !strconcat(OpcodeStr, "pd"), f256mem,
1660 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1661 (bc_v4i64 (v4f64 VR256:$src2))))],
1662 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1663 (memopv4i64 addr:$src2)))], 0>,
1667 // AVX 256-bit packed logical ops forms
1668 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1669 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1670 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1671 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
1673 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1674 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1675 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1676 let isCommutable = 0 in
1677 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
1679 //===----------------------------------------------------------------------===//
1680 // SSE 1 & 2 - Arithmetic Instructions
1681 //===----------------------------------------------------------------------===//
1683 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1686 /// In addition, we also have a special variant of the scalar form here to
1687 /// represent the associated intrinsic operation. This form is unlike the
1688 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1689 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1691 /// These three forms can each be reg+reg or reg+mem.
1694 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1696 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1698 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1699 OpNode, FR32, f32mem, Is2Addr>, XS;
1700 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1701 OpNode, FR64, f64mem, Is2Addr>, XD;
1704 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1706 let mayLoad = 0 in {
1707 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1708 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1709 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1710 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1714 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1716 let mayLoad = 0 in {
1717 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1718 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1719 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1720 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1724 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1726 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1727 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1728 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1729 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1732 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1734 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1735 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1736 SSEPackedSingle, Is2Addr>, TB;
1738 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1739 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1740 SSEPackedDouble, Is2Addr>, TB, OpSize;
1743 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1744 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1745 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1746 SSEPackedSingle, 0>, TB;
1748 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1749 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1750 SSEPackedDouble, 0>, TB, OpSize;
1753 // Binary Arithmetic instructions
1754 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1755 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1756 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1757 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1758 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1759 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1760 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1761 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1763 let isCommutable = 0 in {
1764 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1765 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1766 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1767 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1768 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1769 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1770 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1771 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1772 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1773 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1774 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1775 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1776 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1777 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1778 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1779 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1780 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1781 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1782 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1783 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1786 let Constraints = "$src1 = $dst" in {
1787 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1788 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1789 basic_sse12_fp_binop_s_int<0x58, "add">;
1790 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1791 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1792 basic_sse12_fp_binop_s_int<0x59, "mul">;
1794 let isCommutable = 0 in {
1795 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1796 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1797 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1798 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1799 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1800 basic_sse12_fp_binop_s_int<0x5E, "div">;
1801 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1802 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1803 basic_sse12_fp_binop_s_int<0x5F, "max">,
1804 basic_sse12_fp_binop_p_int<0x5F, "max">;
1805 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1806 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1807 basic_sse12_fp_binop_s_int<0x5D, "min">,
1808 basic_sse12_fp_binop_p_int<0x5D, "min">;
1813 /// In addition, we also have a special variant of the scalar form here to
1814 /// represent the associated intrinsic operation. This form is unlike the
1815 /// plain scalar form, in that it takes an entire vector (instead of a
1816 /// scalar) and leaves the top elements undefined.
1818 /// And, we have a special variant form for a full-vector intrinsic form.
1820 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1821 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1822 SDNode OpNode, Intrinsic F32Int> {
1823 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1824 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1825 [(set FR32:$dst, (OpNode FR32:$src))]>;
1826 // For scalar unary operations, fold a load into the operation
1827 // only in OptForSize mode. It eliminates an instruction, but it also
1828 // eliminates a whole-register clobber (the load), so it introduces a
1829 // partial register update condition.
1830 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1831 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1832 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1833 Requires<[HasSSE1, OptForSize]>;
1834 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1835 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1836 [(set VR128:$dst, (F32Int VR128:$src))]>;
1837 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1838 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1839 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1842 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1843 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
1844 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1845 !strconcat(OpcodeStr,
1846 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1847 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
1848 !strconcat(OpcodeStr,
1849 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1850 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
1851 (ins ssmem:$src1, VR128:$src2),
1852 !strconcat(OpcodeStr,
1853 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1856 /// sse1_fp_unop_p - SSE1 unops in packed form.
1857 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1858 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1859 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1860 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1861 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1862 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1863 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1866 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1867 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1868 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1869 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1870 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1871 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1872 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1873 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1876 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1877 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1878 Intrinsic V4F32Int> {
1879 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1880 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1881 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1882 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1883 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1884 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1887 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1888 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1889 Intrinsic V4F32Int> {
1890 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1891 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1892 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1893 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1894 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1895 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1898 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1899 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1900 SDNode OpNode, Intrinsic F64Int> {
1901 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1902 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1903 [(set FR64:$dst, (OpNode FR64:$src))]>;
1904 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1905 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1906 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1907 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1908 Requires<[HasSSE2, OptForSize]>;
1909 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1910 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1911 [(set VR128:$dst, (F64Int VR128:$src))]>;
1912 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1913 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1914 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1917 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1918 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
1919 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1920 !strconcat(OpcodeStr,
1921 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1922 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
1923 !strconcat(OpcodeStr,
1924 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1925 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
1926 (ins VR128:$src1, sdmem:$src2),
1927 !strconcat(OpcodeStr,
1928 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1931 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1932 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1934 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1935 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1936 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1937 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1938 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1939 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1942 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1943 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1944 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1945 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1946 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1947 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1948 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1949 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1952 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1953 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1954 Intrinsic V2F64Int> {
1955 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1956 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1957 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1958 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1959 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1960 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1963 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1964 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1965 Intrinsic V2F64Int> {
1966 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1967 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1968 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1969 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1970 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1971 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1974 let Predicates = [HasAVX] in {
1976 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
1977 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
1979 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1980 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1981 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1982 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1983 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1984 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1985 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1986 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1989 // Reciprocal approximations. Note that these typically require refinement
1990 // in order to obtain suitable precision.
1991 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
1992 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1993 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1994 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1995 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1997 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
1998 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1999 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2000 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2001 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2004 def : Pat<(f32 (fsqrt FR32:$src)),
2005 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2006 def : Pat<(f32 (fsqrt (load addr:$src))),
2007 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2008 Requires<[HasAVX, OptForSize]>;
2009 def : Pat<(f64 (fsqrt FR64:$src)),
2010 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2011 def : Pat<(f64 (fsqrt (load addr:$src))),
2012 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2013 Requires<[HasAVX, OptForSize]>;
2015 def : Pat<(f32 (X86frsqrt FR32:$src)),
2016 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2017 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2018 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2019 Requires<[HasAVX, OptForSize]>;
2021 def : Pat<(f32 (X86frcp FR32:$src)),
2022 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2023 def : Pat<(f32 (X86frcp (load addr:$src))),
2024 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2025 Requires<[HasAVX, OptForSize]>;
2027 let Predicates = [HasAVX] in {
2028 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2029 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2030 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2031 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2033 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2034 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2036 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2037 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2038 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2039 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2041 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2042 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2044 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2045 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2046 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2047 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2049 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2050 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2052 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2053 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2054 (VRCPSSr (f32 (IMPLICIT_DEF)),
2055 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2057 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2058 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2062 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2063 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2064 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2065 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2066 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2067 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2069 // Reciprocal approximations. Note that these typically require refinement
2070 // in order to obtain suitable precision.
2071 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2072 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2073 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2074 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2075 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2076 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2078 // There is no f64 version of the reciprocal approximation instructions.
2080 //===----------------------------------------------------------------------===//
2081 // SSE 1 & 2 - Non-temporal stores
2082 //===----------------------------------------------------------------------===//
2084 let AddedComplexity = 400 in { // Prefer non-temporal versions
2085 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2086 (ins f128mem:$dst, VR128:$src),
2087 "movntps\t{$src, $dst|$dst, $src}",
2088 [(alignednontemporalstore (v4f32 VR128:$src),
2090 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2091 (ins f128mem:$dst, VR128:$src),
2092 "movntpd\t{$src, $dst|$dst, $src}",
2093 [(alignednontemporalstore (v2f64 VR128:$src),
2095 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2096 (ins f128mem:$dst, VR128:$src),
2097 "movntdq\t{$src, $dst|$dst, $src}",
2098 [(alignednontemporalstore (v2f64 VR128:$src),
2101 let ExeDomain = SSEPackedInt in
2102 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2103 (ins f128mem:$dst, VR128:$src),
2104 "movntdq\t{$src, $dst|$dst, $src}",
2105 [(alignednontemporalstore (v4f32 VR128:$src),
2108 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2109 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
2111 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2112 (ins f256mem:$dst, VR256:$src),
2113 "movntps\t{$src, $dst|$dst, $src}",
2114 [(alignednontemporalstore (v8f32 VR256:$src),
2116 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2117 (ins f256mem:$dst, VR256:$src),
2118 "movntpd\t{$src, $dst|$dst, $src}",
2119 [(alignednontemporalstore (v4f64 VR256:$src),
2121 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2122 (ins f256mem:$dst, VR256:$src),
2123 "movntdq\t{$src, $dst|$dst, $src}",
2124 [(alignednontemporalstore (v4f64 VR256:$src),
2126 let ExeDomain = SSEPackedInt in
2127 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2128 (ins f256mem:$dst, VR256:$src),
2129 "movntdq\t{$src, $dst|$dst, $src}",
2130 [(alignednontemporalstore (v8f32 VR256:$src),
2134 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2135 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2136 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2137 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2138 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2139 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2141 let AddedComplexity = 400 in { // Prefer non-temporal versions
2142 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2143 "movntps\t{$src, $dst|$dst, $src}",
2144 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2145 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2146 "movntpd\t{$src, $dst|$dst, $src}",
2147 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2149 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2150 "movntdq\t{$src, $dst|$dst, $src}",
2151 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2153 let ExeDomain = SSEPackedInt in
2154 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2155 "movntdq\t{$src, $dst|$dst, $src}",
2156 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2158 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2159 (MOVNTDQmr addr:$dst, VR128:$src)>;
2161 // There is no AVX form for instructions below this point
2162 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2163 "movnti{l}\t{$src, $dst|$dst, $src}",
2164 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2165 TB, Requires<[HasSSE2]>;
2166 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2167 "movnti{q}\t{$src, $dst|$dst, $src}",
2168 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2169 TB, Requires<[HasSSE2]>;
2172 //===----------------------------------------------------------------------===//
2173 // SSE 1 & 2 - Misc Instructions (No AVX form)
2174 //===----------------------------------------------------------------------===//
2176 // Prefetch intrinsic.
2177 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2178 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2179 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2180 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2181 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2182 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2183 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2184 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2186 // Load, store, and memory fence
2187 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2188 TB, Requires<[HasSSE1]>;
2189 def : Pat<(X86SFence), (SFENCE)>;
2191 // Alias instructions that map zero vector to pxor / xorp* for sse.
2192 // We set canFoldAsLoad because this can be converted to a constant-pool
2193 // load of an all-zeros value if folding it would be beneficial.
2194 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2195 // JIT implementation, it does not expand the instructions below like
2196 // X86MCInstLower does.
2197 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2198 isCodeGenOnly = 1 in {
2199 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2200 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2201 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2202 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2203 let ExeDomain = SSEPackedInt in
2204 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2205 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2208 // The same as done above but for AVX. The 128-bit versions are the
2209 // same, but re-encoded. The 256-bit does not support PI version, and
2210 // doesn't need it because on sandy bridge the register is set to zero
2211 // at the rename stage without using any execution unit, so SET0PSY
2212 // and SET0PDY can be used for vector int instructions without penalty
2213 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2214 // JIT implementatioan, it does not expand the instructions below like
2215 // X86MCInstLower does.
2216 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2217 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2218 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2219 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2220 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2221 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2222 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2223 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2224 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2225 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2226 let ExeDomain = SSEPackedInt in
2227 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2228 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2231 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2232 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2233 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2235 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2236 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2238 // AVX has no support for 256-bit integer instructions, but since the 128-bit
2239 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
2240 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
2241 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
2242 (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
2244 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
2245 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
2246 (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
2248 //===----------------------------------------------------------------------===//
2249 // SSE 1 & 2 - Load/Store XCSR register
2250 //===----------------------------------------------------------------------===//
2252 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2253 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2254 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2255 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2257 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2258 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2259 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2260 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2262 //===---------------------------------------------------------------------===//
2263 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2264 //===---------------------------------------------------------------------===//
2266 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2268 let neverHasSideEffects = 1 in {
2269 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2270 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2271 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2272 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2274 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2275 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2276 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2277 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2279 let canFoldAsLoad = 1, mayLoad = 1 in {
2280 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2281 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2282 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2283 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2284 let Predicates = [HasAVX] in {
2285 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2286 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2287 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2288 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2292 let mayStore = 1 in {
2293 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2294 (ins i128mem:$dst, VR128:$src),
2295 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2296 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2297 (ins i256mem:$dst, VR256:$src),
2298 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2299 let Predicates = [HasAVX] in {
2300 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2301 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2302 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2303 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2307 let neverHasSideEffects = 1 in
2308 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2309 "movdqa\t{$src, $dst|$dst, $src}", []>;
2311 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2312 "movdqu\t{$src, $dst|$dst, $src}",
2313 []>, XS, Requires<[HasSSE2]>;
2315 let canFoldAsLoad = 1, mayLoad = 1 in {
2316 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2317 "movdqa\t{$src, $dst|$dst, $src}",
2318 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2319 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2320 "movdqu\t{$src, $dst|$dst, $src}",
2321 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2322 XS, Requires<[HasSSE2]>;
2325 let mayStore = 1 in {
2326 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2327 "movdqa\t{$src, $dst|$dst, $src}",
2328 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2329 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2330 "movdqu\t{$src, $dst|$dst, $src}",
2331 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2332 XS, Requires<[HasSSE2]>;
2335 // Intrinsic forms of MOVDQU load and store
2336 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2337 "vmovdqu\t{$src, $dst|$dst, $src}",
2338 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2339 XS, VEX, Requires<[HasAVX]>;
2341 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2342 "movdqu\t{$src, $dst|$dst, $src}",
2343 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2344 XS, Requires<[HasSSE2]>;
2346 } // ExeDomain = SSEPackedInt
2348 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2349 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2350 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2352 //===---------------------------------------------------------------------===//
2353 // SSE2 - Packed Integer Arithmetic Instructions
2354 //===---------------------------------------------------------------------===//
2356 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2358 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2359 bit IsCommutable = 0, bit Is2Addr = 1> {
2360 let isCommutable = IsCommutable in
2361 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2362 (ins VR128:$src1, VR128:$src2),
2364 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2366 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2367 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2368 (ins VR128:$src1, i128mem:$src2),
2370 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2371 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2372 [(set VR128:$dst, (IntId VR128:$src1,
2373 (bitconvert (memopv2i64 addr:$src2))))]>;
2376 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2377 string OpcodeStr, Intrinsic IntId,
2378 Intrinsic IntId2, bit Is2Addr = 1> {
2379 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2380 (ins VR128:$src1, VR128:$src2),
2382 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2383 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2384 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2385 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2386 (ins VR128:$src1, i128mem:$src2),
2388 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2389 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2390 [(set VR128:$dst, (IntId VR128:$src1,
2391 (bitconvert (memopv2i64 addr:$src2))))]>;
2392 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2393 (ins VR128:$src1, i32i8imm:$src2),
2395 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2396 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2397 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2400 /// PDI_binop_rm - Simple SSE2 binary operator.
2401 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2402 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2403 let isCommutable = IsCommutable in
2404 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2405 (ins VR128:$src1, VR128:$src2),
2407 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2408 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2409 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2410 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2411 (ins VR128:$src1, i128mem:$src2),
2413 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2414 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2415 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2416 (bitconvert (memopv2i64 addr:$src2)))))]>;
2419 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2421 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2422 /// to collapse (bitconvert VT to VT) into its operand.
2424 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2425 bit IsCommutable = 0, bit Is2Addr = 1> {
2426 let isCommutable = IsCommutable in
2427 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2428 (ins VR128:$src1, VR128:$src2),
2430 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2431 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2432 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2433 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2434 (ins VR128:$src1, i128mem:$src2),
2436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2438 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2441 } // ExeDomain = SSEPackedInt
2443 // 128-bit Integer Arithmetic
2445 let Predicates = [HasAVX] in {
2446 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2447 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2448 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2449 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2450 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2451 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2452 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2453 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2454 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2457 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2459 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2461 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2463 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2465 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2467 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2469 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2471 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2473 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2475 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2477 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2479 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2481 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2483 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2485 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2487 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2489 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2491 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2493 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2497 let Constraints = "$src1 = $dst" in {
2498 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2499 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2500 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2501 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2502 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2503 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2504 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2505 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2506 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2509 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2510 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2511 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2512 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2513 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2514 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2515 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2516 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2517 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2518 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2519 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2520 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2521 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2522 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2523 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2524 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2525 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2526 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2527 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2529 } // Constraints = "$src1 = $dst"
2531 //===---------------------------------------------------------------------===//
2532 // SSE2 - Packed Integer Logical Instructions
2533 //===---------------------------------------------------------------------===//
2535 let Predicates = [HasAVX] in {
2536 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2537 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2539 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2540 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2542 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2543 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2546 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2547 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2549 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2550 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2552 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2553 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2556 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2557 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2559 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2560 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2563 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2564 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2565 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2567 let ExeDomain = SSEPackedInt in {
2568 let neverHasSideEffects = 1 in {
2569 // 128-bit logical shifts.
2570 def VPSLLDQri : PDIi8<0x73, MRM7r,
2571 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2572 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2574 def VPSRLDQri : PDIi8<0x73, MRM3r,
2575 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2576 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2578 // PSRADQri doesn't exist in SSE[1-3].
2580 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2581 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2582 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2584 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
2586 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2587 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2588 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2589 [(set VR128:$dst, (X86andnp VR128:$src1,
2590 (memopv2i64 addr:$src2)))]>, VEX_4V;
2594 let Constraints = "$src1 = $dst" in {
2595 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2596 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2597 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2598 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2599 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2600 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2602 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2603 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2604 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2605 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2606 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2607 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2609 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2610 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2611 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2612 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2614 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2615 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2616 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2618 let ExeDomain = SSEPackedInt in {
2619 let neverHasSideEffects = 1 in {
2620 // 128-bit logical shifts.
2621 def PSLLDQri : PDIi8<0x73, MRM7r,
2622 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2623 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2624 def PSRLDQri : PDIi8<0x73, MRM3r,
2625 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2626 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2627 // PSRADQri doesn't exist in SSE[1-3].
2629 def PANDNrr : PDI<0xDF, MRMSrcReg,
2630 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2631 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2633 def PANDNrm : PDI<0xDF, MRMSrcMem,
2634 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2635 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2637 } // Constraints = "$src1 = $dst"
2639 let Predicates = [HasAVX] in {
2640 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2641 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2642 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2643 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2644 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2645 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2646 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2647 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2648 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2649 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2651 // Shift up / down and insert zero's.
2652 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2653 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2654 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2655 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2658 let Predicates = [HasSSE2] in {
2659 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2660 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2661 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2662 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2663 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2664 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2665 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2666 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2667 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2668 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2670 // Shift up / down and insert zero's.
2671 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2672 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2673 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2674 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2677 //===---------------------------------------------------------------------===//
2678 // SSE2 - Packed Integer Comparison Instructions
2679 //===---------------------------------------------------------------------===//
2681 let Predicates = [HasAVX] in {
2682 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2684 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2686 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2688 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2690 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2692 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2696 let Constraints = "$src1 = $dst" in {
2697 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2698 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2699 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2700 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2701 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2702 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2703 } // Constraints = "$src1 = $dst"
2705 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2706 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2707 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2708 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2709 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2710 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2711 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2712 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2713 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2714 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2715 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2716 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2718 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2719 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2720 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2721 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2722 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2723 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2724 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2725 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2726 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2727 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2728 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2729 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2731 //===---------------------------------------------------------------------===//
2732 // SSE2 - Packed Integer Pack Instructions
2733 //===---------------------------------------------------------------------===//
2735 let Predicates = [HasAVX] in {
2736 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2738 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2740 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2744 let Constraints = "$src1 = $dst" in {
2745 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2746 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2747 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2748 } // Constraints = "$src1 = $dst"
2750 //===---------------------------------------------------------------------===//
2751 // SSE2 - Packed Integer Shuffle Instructions
2752 //===---------------------------------------------------------------------===//
2754 let ExeDomain = SSEPackedInt in {
2755 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2757 def ri : Ii8<0x70, MRMSrcReg,
2758 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2759 !strconcat(OpcodeStr,
2760 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2761 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2763 def mi : Ii8<0x70, MRMSrcMem,
2764 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2765 !strconcat(OpcodeStr,
2766 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2767 [(set VR128:$dst, (vt (pshuf_frag:$src2
2768 (bc_frag (memopv2i64 addr:$src1)),
2771 } // ExeDomain = SSEPackedInt
2773 let Predicates = [HasAVX] in {
2774 let AddedComplexity = 5 in
2775 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2778 // SSE2 with ImmT == Imm8 and XS prefix.
2779 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2782 // SSE2 with ImmT == Imm8 and XD prefix.
2783 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2787 let Predicates = [HasSSE2] in {
2788 let AddedComplexity = 5 in
2789 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2791 // SSE2 with ImmT == Imm8 and XS prefix.
2792 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2794 // SSE2 with ImmT == Imm8 and XD prefix.
2795 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2798 //===---------------------------------------------------------------------===//
2799 // SSE2 - Packed Integer Unpack Instructions
2800 //===---------------------------------------------------------------------===//
2802 let ExeDomain = SSEPackedInt in {
2803 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2804 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
2805 def rr : PDI<opc, MRMSrcReg,
2806 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2808 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2809 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2810 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
2811 def rm : PDI<opc, MRMSrcMem,
2812 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2814 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2815 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2816 [(set VR128:$dst, (OpNode VR128:$src1,
2817 (bc_frag (memopv2i64
2821 let Predicates = [HasAVX] in {
2822 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
2823 bc_v16i8, 0>, VEX_4V;
2824 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
2825 bc_v8i16, 0>, VEX_4V;
2826 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
2827 bc_v4i32, 0>, VEX_4V;
2829 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2830 /// knew to collapse (bitconvert VT to VT) into its operand.
2831 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2832 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2833 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2834 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
2835 VR128:$src2)))]>, VEX_4V;
2836 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2837 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2838 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2839 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
2840 (memopv2i64 addr:$src2))))]>, VEX_4V;
2842 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
2843 bc_v16i8, 0>, VEX_4V;
2844 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
2845 bc_v8i16, 0>, VEX_4V;
2846 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
2847 bc_v4i32, 0>, VEX_4V;
2849 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2850 /// knew to collapse (bitconvert VT to VT) into its operand.
2851 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2852 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2853 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2854 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
2855 VR128:$src2)))]>, VEX_4V;
2856 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2857 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2858 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2859 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
2860 (memopv2i64 addr:$src2))))]>, VEX_4V;
2863 let Constraints = "$src1 = $dst" in {
2864 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
2865 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
2866 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
2868 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2869 /// knew to collapse (bitconvert VT to VT) into its operand.
2870 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2871 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2872 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2874 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
2875 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2876 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2877 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2879 (v2i64 (X86Punpcklqdq VR128:$src1,
2880 (memopv2i64 addr:$src2))))]>;
2882 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
2883 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
2884 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
2886 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2887 /// knew to collapse (bitconvert VT to VT) into its operand.
2888 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2889 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2890 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2892 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
2893 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2894 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2895 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2897 (v2i64 (X86Punpckhqdq VR128:$src1,
2898 (memopv2i64 addr:$src2))))]>;
2901 } // ExeDomain = SSEPackedInt
2903 //===---------------------------------------------------------------------===//
2904 // SSE2 - Packed Integer Extract and Insert
2905 //===---------------------------------------------------------------------===//
2907 let ExeDomain = SSEPackedInt in {
2908 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2909 def rri : Ii8<0xC4, MRMSrcReg,
2910 (outs VR128:$dst), (ins VR128:$src1,
2911 GR32:$src2, i32i8imm:$src3),
2913 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2914 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2916 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2917 def rmi : Ii8<0xC4, MRMSrcMem,
2918 (outs VR128:$dst), (ins VR128:$src1,
2919 i16mem:$src2, i32i8imm:$src3),
2921 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2922 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2924 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2929 let Predicates = [HasAVX] in
2930 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2931 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2932 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2933 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2934 imm:$src2))]>, OpSize, VEX;
2935 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2936 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2937 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2938 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2942 let Predicates = [HasAVX] in {
2943 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2944 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2945 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2946 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2947 []>, OpSize, VEX_4V;
2950 let Constraints = "$src1 = $dst" in
2951 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2953 } // ExeDomain = SSEPackedInt
2955 //===---------------------------------------------------------------------===//
2956 // SSE2 - Packed Mask Creation
2957 //===---------------------------------------------------------------------===//
2959 let ExeDomain = SSEPackedInt in {
2961 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2962 "pmovmskb\t{$src, $dst|$dst, $src}",
2963 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2964 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2965 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2966 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2967 "pmovmskb\t{$src, $dst|$dst, $src}",
2968 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2970 } // ExeDomain = SSEPackedInt
2972 //===---------------------------------------------------------------------===//
2973 // SSE2 - Conditional Store
2974 //===---------------------------------------------------------------------===//
2976 let ExeDomain = SSEPackedInt in {
2979 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2980 (ins VR128:$src, VR128:$mask),
2981 "maskmovdqu\t{$mask, $src|$src, $mask}",
2982 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2984 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2985 (ins VR128:$src, VR128:$mask),
2986 "maskmovdqu\t{$mask, $src|$src, $mask}",
2987 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2990 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2991 "maskmovdqu\t{$mask, $src|$src, $mask}",
2992 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2994 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2995 "maskmovdqu\t{$mask, $src|$src, $mask}",
2996 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2998 } // ExeDomain = SSEPackedInt
3000 //===---------------------------------------------------------------------===//
3001 // SSE2 - Move Doubleword
3002 //===---------------------------------------------------------------------===//
3004 //===---------------------------------------------------------------------===//
3005 // Move Int Doubleword to Packed Double Int
3007 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3008 "movd\t{$src, $dst|$dst, $src}",
3010 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
3011 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3012 "movd\t{$src, $dst|$dst, $src}",
3014 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
3016 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3017 "mov{d|q}\t{$src, $dst|$dst, $src}",
3019 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
3020 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3021 "mov{d|q}\t{$src, $dst|$dst, $src}",
3022 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
3024 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3025 "movd\t{$src, $dst|$dst, $src}",
3027 (v4i32 (scalar_to_vector GR32:$src)))]>;
3028 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3029 "movd\t{$src, $dst|$dst, $src}",
3031 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
3032 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3033 "mov{d|q}\t{$src, $dst|$dst, $src}",
3035 (v2i64 (scalar_to_vector GR64:$src)))]>;
3036 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3037 "mov{d|q}\t{$src, $dst|$dst, $src}",
3038 [(set FR64:$dst, (bitconvert GR64:$src))]>;
3040 //===---------------------------------------------------------------------===//
3041 // Move Int Doubleword to Single Scalar
3043 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3044 "movd\t{$src, $dst|$dst, $src}",
3045 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
3047 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3048 "movd\t{$src, $dst|$dst, $src}",
3049 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
3051 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3052 "movd\t{$src, $dst|$dst, $src}",
3053 [(set FR32:$dst, (bitconvert GR32:$src))]>;
3055 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3056 "movd\t{$src, $dst|$dst, $src}",
3057 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
3059 //===---------------------------------------------------------------------===//
3060 // Move Packed Doubleword Int to Packed Double Int
3062 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3063 "movd\t{$src, $dst|$dst, $src}",
3064 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3066 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
3067 (ins i32mem:$dst, VR128:$src),
3068 "movd\t{$src, $dst|$dst, $src}",
3069 [(store (i32 (vector_extract (v4i32 VR128:$src),
3070 (iPTR 0))), addr:$dst)]>, VEX;
3071 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3072 "movd\t{$src, $dst|$dst, $src}",
3073 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3075 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3076 "movd\t{$src, $dst|$dst, $src}",
3077 [(store (i32 (vector_extract (v4i32 VR128:$src),
3078 (iPTR 0))), addr:$dst)]>;
3080 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3081 "mov{d|q}\t{$src, $dst|$dst, $src}",
3082 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
3084 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
3085 "movq\t{$src, $dst|$dst, $src}",
3086 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
3088 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
3089 "mov{d|q}\t{$src, $dst|$dst, $src}",
3090 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3091 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3092 "movq\t{$src, $dst|$dst, $src}",
3093 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3095 //===---------------------------------------------------------------------===//
3096 // Move Scalar Single to Double Int
3098 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3099 "movd\t{$src, $dst|$dst, $src}",
3100 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3101 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3102 "movd\t{$src, $dst|$dst, $src}",
3103 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3104 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3105 "movd\t{$src, $dst|$dst, $src}",
3106 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3107 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3108 "movd\t{$src, $dst|$dst, $src}",
3109 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3111 //===---------------------------------------------------------------------===//
3112 // Patterns and instructions to describe movd/movq to XMM register zero-extends
3114 let AddedComplexity = 15 in {
3115 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3116 "movd\t{$src, $dst|$dst, $src}",
3117 [(set VR128:$dst, (v4i32 (X86vzmovl
3118 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3120 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3121 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3122 [(set VR128:$dst, (v2i64 (X86vzmovl
3123 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3126 let AddedComplexity = 15 in {
3127 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3128 "movd\t{$src, $dst|$dst, $src}",
3129 [(set VR128:$dst, (v4i32 (X86vzmovl
3130 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3131 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3132 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3133 [(set VR128:$dst, (v2i64 (X86vzmovl
3134 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3137 let AddedComplexity = 20 in {
3138 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3139 "movd\t{$src, $dst|$dst, $src}",
3141 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3142 (loadi32 addr:$src))))))]>,
3144 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3145 "movd\t{$src, $dst|$dst, $src}",
3147 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3148 (loadi32 addr:$src))))))]>;
3150 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3151 (MOVZDI2PDIrm addr:$src)>;
3152 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3153 (MOVZDI2PDIrm addr:$src)>;
3154 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3155 (MOVZDI2PDIrm addr:$src)>;
3158 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
3159 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
3160 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
3161 (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
3162 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
3163 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
3164 (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
3165 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
3167 // These are the correct encodings of the instructions so that we know how to
3168 // read correct assembly, even though we continue to emit the wrong ones for
3169 // compatibility with Darwin's buggy assembler.
3170 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3171 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
3172 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3173 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
3174 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3175 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
3176 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3177 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
3178 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3179 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3180 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3181 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3183 //===---------------------------------------------------------------------===//
3184 // SSE2 - Move Quadword
3185 //===---------------------------------------------------------------------===//
3187 //===---------------------------------------------------------------------===//
3188 // Move Quadword Int to Packed Quadword Int
3190 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3191 "vmovq\t{$src, $dst|$dst, $src}",
3193 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3194 VEX, Requires<[HasAVX]>;
3195 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3196 "movq\t{$src, $dst|$dst, $src}",
3198 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3199 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3201 //===---------------------------------------------------------------------===//
3202 // Move Packed Quadword Int to Quadword Int
3204 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3205 "movq\t{$src, $dst|$dst, $src}",
3206 [(store (i64 (vector_extract (v2i64 VR128:$src),
3207 (iPTR 0))), addr:$dst)]>, VEX;
3208 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3209 "movq\t{$src, $dst|$dst, $src}",
3210 [(store (i64 (vector_extract (v2i64 VR128:$src),
3211 (iPTR 0))), addr:$dst)]>;
3213 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3214 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3216 //===---------------------------------------------------------------------===//
3217 // Store / copy lower 64-bits of a XMM register.
3219 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3220 "movq\t{$src, $dst|$dst, $src}",
3221 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3222 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3223 "movq\t{$src, $dst|$dst, $src}",
3224 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3226 let AddedComplexity = 20 in
3227 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3228 "vmovq\t{$src, $dst|$dst, $src}",
3230 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3231 (loadi64 addr:$src))))))]>,
3232 XS, VEX, Requires<[HasAVX]>;
3234 let AddedComplexity = 20 in {
3235 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3236 "movq\t{$src, $dst|$dst, $src}",
3238 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3239 (loadi64 addr:$src))))))]>,
3240 XS, Requires<[HasSSE2]>;
3242 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3243 (MOVZQI2PQIrm addr:$src)>;
3244 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3245 (MOVZQI2PQIrm addr:$src)>;
3246 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3249 //===---------------------------------------------------------------------===//
3250 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3251 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3253 let AddedComplexity = 15 in
3254 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3255 "vmovq\t{$src, $dst|$dst, $src}",
3256 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3257 XS, VEX, Requires<[HasAVX]>;
3258 let AddedComplexity = 15 in
3259 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3260 "movq\t{$src, $dst|$dst, $src}",
3261 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3262 XS, Requires<[HasSSE2]>;
3264 let AddedComplexity = 20 in
3265 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3266 "vmovq\t{$src, $dst|$dst, $src}",
3267 [(set VR128:$dst, (v2i64 (X86vzmovl
3268 (loadv2i64 addr:$src))))]>,
3269 XS, VEX, Requires<[HasAVX]>;
3270 let AddedComplexity = 20 in {
3271 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3272 "movq\t{$src, $dst|$dst, $src}",
3273 [(set VR128:$dst, (v2i64 (X86vzmovl
3274 (loadv2i64 addr:$src))))]>,
3275 XS, Requires<[HasSSE2]>;
3277 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3278 (MOVZPQILo2PQIrm addr:$src)>;
3281 // Instructions to match in the assembler
3282 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3283 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3284 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3285 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3286 // Recognize "movd" with GR64 destination, but encode as a "movq"
3287 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3288 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3290 // Instructions for the disassembler
3291 // xr = XMM register
3294 let Predicates = [HasAVX] in
3295 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3296 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3297 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3298 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3300 //===---------------------------------------------------------------------===//
3301 // SSE2 - Misc Instructions
3302 //===---------------------------------------------------------------------===//
3305 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3306 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3307 TB, Requires<[HasSSE2]>;
3309 // Load, store, and memory fence
3310 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3311 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3312 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3313 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3314 def : Pat<(X86LFence), (LFENCE)>;
3315 def : Pat<(X86MFence), (MFENCE)>;
3318 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3319 // was introduced with SSE2, it's backward compatible.
3320 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3322 // Alias instructions that map zero vector to pxor / xorp* for sse.
3323 // We set canFoldAsLoad because this can be converted to a constant-pool
3324 // load of an all-ones value if folding it would be beneficial.
3325 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
3326 // JIT implementation, it does not expand the instructions below like
3327 // X86MCInstLower does.
3328 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3329 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3330 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3331 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3332 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3333 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
3334 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3335 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
3337 //===---------------------------------------------------------------------===//
3338 // SSE3 - Conversion Instructions
3339 //===---------------------------------------------------------------------===//
3341 // Convert Packed Double FP to Packed DW Integers
3342 let Predicates = [HasAVX] in {
3343 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3344 // register, but the same isn't true when using memory operands instead.
3345 // Provide other assembly rr and rm forms to address this explicitly.
3346 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3347 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3348 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3349 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3352 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3353 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3354 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3355 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3358 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3359 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3360 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3361 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3364 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3365 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3366 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3367 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3369 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
3370 (VCVTPD2DQYrr VR256:$src)>;
3371 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
3372 (VCVTPD2DQYrm addr:$src)>;
3374 // Convert Packed DW Integers to Packed Double FP
3375 let Predicates = [HasAVX] in {
3376 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3377 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3378 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3379 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3380 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3381 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3382 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3383 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3386 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3387 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3388 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3389 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3391 // AVX 256-bit register conversion intrinsics
3392 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3393 (VCVTDQ2PDYrr VR128:$src)>;
3394 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3395 (VCVTDQ2PDYrm addr:$src)>;
3397 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3398 (VCVTPD2DQYrr VR256:$src)>;
3399 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3400 (VCVTPD2DQYrm addr:$src)>;
3402 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
3403 (VCVTDQ2PDYrr VR128:$src)>;
3404 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
3405 (VCVTDQ2PDYrm addr:$src)>;
3407 //===---------------------------------------------------------------------===//
3408 // SSE3 - Move Instructions
3409 //===---------------------------------------------------------------------===//
3411 //===---------------------------------------------------------------------===//
3412 // Replicate Single FP - MOVSHDUP and MOVSLDUP
3414 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
3415 ValueType vt, RegisterClass RC, PatFrag mem_frag,
3416 X86MemOperand x86memop> {
3417 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
3418 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3419 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
3420 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
3421 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3422 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
3425 let Predicates = [HasAVX] in {
3426 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3427 v4f32, VR128, memopv4f32, f128mem>, VEX;
3428 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3429 v4f32, VR128, memopv4f32, f128mem>, VEX;
3430 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3431 v8f32, VR256, memopv8f32, f256mem>, VEX;
3432 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3433 v8f32, VR256, memopv8f32, f256mem>, VEX;
3435 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
3436 memopv4f32, f128mem>;
3437 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
3438 memopv4f32, f128mem>;
3440 let Predicates = [HasSSE3] in {
3441 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3442 (MOVSHDUPrr VR128:$src)>;
3443 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3444 (MOVSHDUPrm addr:$src)>;
3445 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3446 (MOVSLDUPrr VR128:$src)>;
3447 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3448 (MOVSLDUPrm addr:$src)>;
3451 let Predicates = [HasAVX] in {
3452 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3453 (VMOVSHDUPrr VR128:$src)>;
3454 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3455 (VMOVSHDUPrm addr:$src)>;
3456 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3457 (VMOVSLDUPrr VR128:$src)>;
3458 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3459 (VMOVSLDUPrm addr:$src)>;
3460 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
3461 (VMOVSHDUPYrr VR256:$src)>;
3462 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
3463 (VMOVSHDUPYrm addr:$src)>;
3464 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
3465 (VMOVSLDUPYrr VR256:$src)>;
3466 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
3467 (VMOVSLDUPYrm addr:$src)>;
3470 //===---------------------------------------------------------------------===//
3471 // Replicate Double FP - MOVDDUP
3473 multiclass sse3_replicate_dfp<string OpcodeStr> {
3474 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3475 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3476 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3477 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3478 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3480 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3484 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3485 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3486 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3488 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3489 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3493 let Predicates = [HasAVX] in {
3494 // FIXME: Merge above classes when we have patterns for the ymm version
3495 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3496 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3498 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3500 // Move Unaligned Integer
3501 let Predicates = [HasAVX] in {
3502 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3503 "vlddqu\t{$src, $dst|$dst, $src}",
3504 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3505 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3506 "vlddqu\t{$src, $dst|$dst, $src}",
3507 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3509 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3510 "lddqu\t{$src, $dst|$dst, $src}",
3511 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3513 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3515 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3517 // Several Move patterns
3518 let AddedComplexity = 5 in {
3519 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3520 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3521 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3522 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3523 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3524 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3525 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3526 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3529 //===---------------------------------------------------------------------===//
3530 // SSE3 - Arithmetic
3531 //===---------------------------------------------------------------------===//
3533 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3534 X86MemOperand x86memop, bit Is2Addr = 1> {
3535 def rr : I<0xD0, MRMSrcReg,
3536 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3538 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3539 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3540 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3541 def rm : I<0xD0, MRMSrcMem,
3542 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3544 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3545 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3546 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3549 let Predicates = [HasAVX],
3550 ExeDomain = SSEPackedDouble in {
3551 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3552 f128mem, 0>, TB, XD, VEX_4V;
3553 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3554 f128mem, 0>, TB, OpSize, VEX_4V;
3555 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3556 f256mem, 0>, TB, XD, VEX_4V;
3557 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3558 f256mem, 0>, TB, OpSize, VEX_4V;
3560 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3561 ExeDomain = SSEPackedDouble in {
3562 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3564 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3565 f128mem>, TB, OpSize;
3568 //===---------------------------------------------------------------------===//
3569 // SSE3 Instructions
3570 //===---------------------------------------------------------------------===//
3573 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3574 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3575 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3577 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3578 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3579 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3581 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3583 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3584 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3585 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3587 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3588 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3589 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3591 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3592 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3593 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3595 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3597 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3598 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3599 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3602 let Predicates = [HasAVX] in {
3603 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3604 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3605 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3606 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3607 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3608 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3609 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3610 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3611 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3612 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3613 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3614 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3615 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3616 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3617 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3618 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3621 let Constraints = "$src1 = $dst" in {
3622 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3623 int_x86_sse3_hadd_ps>;
3624 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3625 int_x86_sse3_hadd_pd>;
3626 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3627 int_x86_sse3_hsub_ps>;
3628 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3629 int_x86_sse3_hsub_pd>;
3632 //===---------------------------------------------------------------------===//
3633 // SSSE3 - Packed Absolute Instructions
3634 //===---------------------------------------------------------------------===//
3637 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3638 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3639 PatFrag mem_frag128, Intrinsic IntId128> {
3640 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3642 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3643 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3646 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3648 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3651 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3654 let Predicates = [HasAVX] in {
3655 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3656 int_x86_ssse3_pabs_b_128>, VEX;
3657 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3658 int_x86_ssse3_pabs_w_128>, VEX;
3659 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3660 int_x86_ssse3_pabs_d_128>, VEX;
3663 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3664 int_x86_ssse3_pabs_b_128>;
3665 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3666 int_x86_ssse3_pabs_w_128>;
3667 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3668 int_x86_ssse3_pabs_d_128>;
3670 //===---------------------------------------------------------------------===//
3671 // SSSE3 - Packed Binary Operator Instructions
3672 //===---------------------------------------------------------------------===//
3674 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3675 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3676 PatFrag mem_frag128, Intrinsic IntId128,
3678 let isCommutable = 1 in
3679 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3680 (ins VR128:$src1, VR128:$src2),
3682 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3683 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3684 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3686 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3687 (ins VR128:$src1, i128mem:$src2),
3689 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3690 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3692 (IntId128 VR128:$src1,
3693 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3696 let Predicates = [HasAVX] in {
3697 let isCommutable = 0 in {
3698 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3699 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3700 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3701 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3702 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3703 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3704 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3705 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3706 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3707 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3708 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3709 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3710 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3711 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3712 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3713 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3714 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3715 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3716 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3717 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3718 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3719 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3721 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3722 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3725 // None of these have i8 immediate fields.
3726 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3727 let isCommutable = 0 in {
3728 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3729 int_x86_ssse3_phadd_w_128>;
3730 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3731 int_x86_ssse3_phadd_d_128>;
3732 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3733 int_x86_ssse3_phadd_sw_128>;
3734 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3735 int_x86_ssse3_phsub_w_128>;
3736 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3737 int_x86_ssse3_phsub_d_128>;
3738 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3739 int_x86_ssse3_phsub_sw_128>;
3740 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3741 int_x86_ssse3_pmadd_ub_sw_128>;
3742 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3743 int_x86_ssse3_pshuf_b_128>;
3744 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3745 int_x86_ssse3_psign_b_128>;
3746 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3747 int_x86_ssse3_psign_w_128>;
3748 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3749 int_x86_ssse3_psign_d_128>;
3751 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3752 int_x86_ssse3_pmul_hr_sw_128>;
3755 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3756 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3757 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3758 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3760 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3761 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3762 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3763 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3764 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3765 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3767 //===---------------------------------------------------------------------===//
3768 // SSSE3 - Packed Align Instruction Patterns
3769 //===---------------------------------------------------------------------===//
3771 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3772 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3773 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3775 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3777 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3779 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3780 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3782 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3784 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3788 let Predicates = [HasAVX] in
3789 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3790 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
3791 defm PALIGN : ssse3_palign<"palignr">;
3793 let Predicates = [HasSSSE3] in {
3794 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3795 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3796 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3797 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3798 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3799 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3800 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3801 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3804 let Predicates = [HasAVX] in {
3805 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3806 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3807 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3808 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3809 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3810 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3811 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
3812 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
3815 //===---------------------------------------------------------------------===//
3816 // SSSE3 Misc Instructions
3817 //===---------------------------------------------------------------------===//
3819 // Thread synchronization
3820 let usesCustomInserter = 1 in {
3821 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3822 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3823 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3824 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3827 let Uses = [EAX, ECX, EDX] in
3828 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3829 Requires<[HasSSE3]>;
3830 let Uses = [ECX, EAX] in
3831 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3832 Requires<[HasSSE3]>;
3834 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
3835 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
3837 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
3838 Requires<[In32BitMode]>;
3839 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
3840 Requires<[In64BitMode]>;
3842 //===---------------------------------------------------------------------===//
3843 // Non-Instruction Patterns
3844 //===---------------------------------------------------------------------===//
3846 // extload f32 -> f64. This matches load+fextend because we have a hack in
3847 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3849 // Since these loads aren't folded into the fextend, we have to match it
3851 let Predicates = [HasSSE2] in
3852 def : Pat<(fextend (loadf32 addr:$src)),
3853 (CVTSS2SDrm addr:$src)>;
3855 // Bitcasts between 128-bit vector types. Return the original type since
3856 // no instruction is needed for the conversion
3857 let Predicates = [HasXMMInt] in {
3858 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3859 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3860 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3861 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3862 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3863 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3864 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3865 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3866 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3867 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3868 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3869 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3870 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3871 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3872 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3873 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3874 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3875 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3876 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3877 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3878 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3879 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3880 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3881 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3882 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3883 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3884 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3885 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3886 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3887 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3890 // Bitcasts between 256-bit vector types. Return the original type since
3891 // no instruction is needed for the conversion
3892 let Predicates = [HasAVX] in {
3893 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3894 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
3895 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
3896 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
3897 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
3898 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
3899 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
3900 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
3901 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
3902 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
3903 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
3904 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
3905 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
3906 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
3907 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
3908 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
3909 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
3910 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
3911 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
3912 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
3913 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
3914 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
3915 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
3916 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
3917 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
3918 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
3919 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
3920 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
3921 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
3922 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
3925 // Move scalar to XMM zero-extended
3926 // movd to XMM register zero-extends
3927 let AddedComplexity = 15 in {
3928 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3929 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3930 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3931 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3932 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3933 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3934 (MOVSSrr (v4f32 (V_SET0PS)),
3935 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3936 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3937 (MOVSSrr (v4i32 (V_SET0PI)),
3938 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3941 // Splat v2f64 / v2i64
3942 let AddedComplexity = 10 in {
3943 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3944 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3945 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3946 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3949 // Special unary SHUFPSrri case.
3950 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3951 (SHUFPSrri VR128:$src1, VR128:$src1,
3952 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3953 let AddedComplexity = 5 in
3954 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3955 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3956 Requires<[HasSSE2]>;
3957 // Special unary SHUFPDrri case.
3958 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3959 (SHUFPDrri VR128:$src1, VR128:$src1,
3960 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3961 Requires<[HasSSE2]>;
3962 // Special unary SHUFPDrri case.
3963 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3964 (SHUFPDrri VR128:$src1, VR128:$src1,
3965 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3966 Requires<[HasSSE2]>;
3967 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3968 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3969 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3970 Requires<[HasSSE2]>;
3972 // Special binary v4i32 shuffle cases with SHUFPS.
3973 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3974 (SHUFPSrri VR128:$src1, VR128:$src2,
3975 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3976 Requires<[HasSSE2]>;
3977 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3978 (SHUFPSrmi VR128:$src1, addr:$src2,
3979 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3980 Requires<[HasSSE2]>;
3981 // Special binary v2i64 shuffle cases using SHUFPDrri.
3982 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3983 (SHUFPDrri VR128:$src1, VR128:$src2,
3984 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3985 Requires<[HasSSE2]>;
3987 let AddedComplexity = 20 in {
3988 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3989 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3990 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3992 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3993 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3994 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3996 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3997 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3998 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3999 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
4000 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
4003 let AddedComplexity = 20 in {
4004 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
4005 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
4006 (MOVLPSrm VR128:$src1, addr:$src2)>;
4007 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
4008 (MOVLPDrm VR128:$src1, addr:$src2)>;
4009 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
4010 (MOVLPSrm VR128:$src1, addr:$src2)>;
4011 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
4012 (MOVLPDrm VR128:$src1, addr:$src2)>;
4015 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
4016 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4017 (MOVLPSmr addr:$src1, VR128:$src2)>;
4018 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4019 (MOVLPDmr addr:$src1, VR128:$src2)>;
4020 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
4022 (MOVLPSmr addr:$src1, VR128:$src2)>;
4023 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4024 (MOVLPDmr addr:$src1, VR128:$src2)>;
4026 let AddedComplexity = 15 in {
4027 // Setting the lowest element in the vector.
4028 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
4029 (MOVSSrr (v4i32 VR128:$src1),
4030 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
4031 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
4032 (MOVSDrr (v2i64 VR128:$src1),
4033 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
4035 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
4036 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
4037 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4038 Requires<[HasSSE2]>;
4039 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
4040 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4041 Requires<[HasSSE2]>;
4044 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
4045 // fall back to this for SSE1)
4046 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
4047 (SHUFPSrri VR128:$src2, VR128:$src1,
4048 (SHUFFLE_get_shuf_imm VR128:$src3))>;
4050 // Set lowest element and zero upper elements.
4051 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4052 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
4054 // Use movaps / movups for SSE integer load / store (one byte shorter).
4055 // The instructions selected below are then converted to MOVDQA/MOVDQU
4056 // during the SSE domain pass.
4057 let Predicates = [HasSSE1] in {
4058 def : Pat<(alignedloadv4i32 addr:$src),
4059 (MOVAPSrm addr:$src)>;
4060 def : Pat<(loadv4i32 addr:$src),
4061 (MOVUPSrm addr:$src)>;
4062 def : Pat<(alignedloadv2i64 addr:$src),
4063 (MOVAPSrm addr:$src)>;
4064 def : Pat<(loadv2i64 addr:$src),
4065 (MOVUPSrm addr:$src)>;
4067 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4068 (MOVAPSmr addr:$dst, VR128:$src)>;
4069 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4070 (MOVAPSmr addr:$dst, VR128:$src)>;
4071 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4072 (MOVAPSmr addr:$dst, VR128:$src)>;
4073 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4074 (MOVAPSmr addr:$dst, VR128:$src)>;
4075 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4076 (MOVUPSmr addr:$dst, VR128:$src)>;
4077 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4078 (MOVUPSmr addr:$dst, VR128:$src)>;
4079 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4080 (MOVUPSmr addr:$dst, VR128:$src)>;
4081 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4082 (MOVUPSmr addr:$dst, VR128:$src)>;
4085 // Use vmovaps/vmovups for AVX integer load/store.
4086 let Predicates = [HasAVX] in {
4087 // 128-bit load/store
4088 def : Pat<(alignedloadv4i32 addr:$src),
4089 (VMOVAPSrm addr:$src)>;
4090 def : Pat<(loadv4i32 addr:$src),
4091 (VMOVUPSrm addr:$src)>;
4092 def : Pat<(alignedloadv2i64 addr:$src),
4093 (VMOVAPSrm addr:$src)>;
4094 def : Pat<(loadv2i64 addr:$src),
4095 (VMOVUPSrm addr:$src)>;
4097 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4098 (VMOVAPSmr addr:$dst, VR128:$src)>;
4099 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4100 (VMOVAPSmr addr:$dst, VR128:$src)>;
4101 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4102 (VMOVAPSmr addr:$dst, VR128:$src)>;
4103 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4104 (VMOVAPSmr addr:$dst, VR128:$src)>;
4105 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4106 (VMOVUPSmr addr:$dst, VR128:$src)>;
4107 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4108 (VMOVUPSmr addr:$dst, VR128:$src)>;
4109 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4110 (VMOVUPSmr addr:$dst, VR128:$src)>;
4111 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4112 (VMOVUPSmr addr:$dst, VR128:$src)>;
4114 // 256-bit load/store
4115 def : Pat<(alignedloadv4i64 addr:$src),
4116 (VMOVAPSYrm addr:$src)>;
4117 def : Pat<(loadv4i64 addr:$src),
4118 (VMOVUPSYrm addr:$src)>;
4119 def : Pat<(alignedloadv8i32 addr:$src),
4120 (VMOVAPSYrm addr:$src)>;
4121 def : Pat<(loadv8i32 addr:$src),
4122 (VMOVUPSYrm addr:$src)>;
4123 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
4124 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4125 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
4126 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4127 def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
4128 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4129 def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
4130 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4131 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
4132 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4133 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
4134 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4135 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
4136 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4137 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
4138 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4141 //===----------------------------------------------------------------------===//
4142 // SSE4.1 - Packed Move with Sign/Zero Extend
4143 //===----------------------------------------------------------------------===//
4145 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4146 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4147 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4148 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4150 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4151 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4153 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4157 let Predicates = [HasAVX] in {
4158 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4160 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4162 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4164 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4166 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4168 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4172 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4173 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4174 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4175 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4176 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4177 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4179 // Common patterns involving scalar load.
4180 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4181 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4182 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4183 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4185 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4186 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4187 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4188 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4190 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4191 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4192 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4193 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4195 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4196 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4197 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4198 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4200 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4201 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4202 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4203 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4205 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4206 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4207 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4208 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4211 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4212 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4213 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4214 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4216 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4217 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4219 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4223 let Predicates = [HasAVX] in {
4224 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4226 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4228 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4230 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4234 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4235 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4236 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4237 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4239 // Common patterns involving scalar load
4240 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4241 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4242 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4243 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4245 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4246 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4247 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4248 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4251 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4252 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4253 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4254 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4256 // Expecting a i16 load any extended to i32 value.
4257 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4258 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4259 [(set VR128:$dst, (IntId (bitconvert
4260 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4264 let Predicates = [HasAVX] in {
4265 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4267 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4270 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4271 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4273 // Common patterns involving scalar load
4274 def : Pat<(int_x86_sse41_pmovsxbq
4275 (bitconvert (v4i32 (X86vzmovl
4276 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4277 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4279 def : Pat<(int_x86_sse41_pmovzxbq
4280 (bitconvert (v4i32 (X86vzmovl
4281 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4282 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4284 //===----------------------------------------------------------------------===//
4285 // SSE4.1 - Extract Instructions
4286 //===----------------------------------------------------------------------===//
4288 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4289 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4290 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4291 (ins VR128:$src1, i32i8imm:$src2),
4292 !strconcat(OpcodeStr,
4293 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4294 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4296 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4297 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4298 !strconcat(OpcodeStr,
4299 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4302 // There's an AssertZext in the way of writing the store pattern
4303 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4306 let Predicates = [HasAVX] in {
4307 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4308 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4309 (ins VR128:$src1, i32i8imm:$src2),
4310 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4313 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4316 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4317 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4318 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4319 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4320 !strconcat(OpcodeStr,
4321 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4324 // There's an AssertZext in the way of writing the store pattern
4325 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4328 let Predicates = [HasAVX] in
4329 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4331 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4334 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4335 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4336 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4337 (ins VR128:$src1, i32i8imm:$src2),
4338 !strconcat(OpcodeStr,
4339 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4341 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4342 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4343 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4344 !strconcat(OpcodeStr,
4345 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4346 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4347 addr:$dst)]>, OpSize;
4350 let Predicates = [HasAVX] in
4351 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4353 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4355 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4356 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4357 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4358 (ins VR128:$src1, i32i8imm:$src2),
4359 !strconcat(OpcodeStr,
4360 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4362 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4363 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4364 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4365 !strconcat(OpcodeStr,
4366 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4367 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4368 addr:$dst)]>, OpSize, REX_W;
4371 let Predicates = [HasAVX] in
4372 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4374 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4376 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4378 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4379 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4380 (ins VR128:$src1, i32i8imm:$src2),
4381 !strconcat(OpcodeStr,
4382 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4384 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4386 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4387 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4388 !strconcat(OpcodeStr,
4389 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4390 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4391 addr:$dst)]>, OpSize;
4394 let Predicates = [HasAVX] in {
4395 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4396 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4397 (ins VR128:$src1, i32i8imm:$src2),
4398 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4401 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4403 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4404 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4407 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4408 Requires<[HasSSE41]>;
4410 //===----------------------------------------------------------------------===//
4411 // SSE4.1 - Insert Instructions
4412 //===----------------------------------------------------------------------===//
4414 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4415 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4416 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4418 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4420 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4422 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4423 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4424 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4426 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4428 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4430 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4431 imm:$src3))]>, OpSize;
4434 let Predicates = [HasAVX] in
4435 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4436 let Constraints = "$src1 = $dst" in
4437 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4439 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4440 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4441 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4443 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4445 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4447 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4449 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4450 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4452 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4454 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4456 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4457 imm:$src3)))]>, OpSize;
4460 let Predicates = [HasAVX] in
4461 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4462 let Constraints = "$src1 = $dst" in
4463 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4465 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4466 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4467 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4469 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4471 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4473 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4475 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4476 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4478 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4480 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4482 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4483 imm:$src3)))]>, OpSize;
4486 let Predicates = [HasAVX] in
4487 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4488 let Constraints = "$src1 = $dst" in
4489 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4491 // insertps has a few different modes, there's the first two here below which
4492 // are optimized inserts that won't zero arbitrary elements in the destination
4493 // vector. The next one matches the intrinsic and could zero arbitrary elements
4494 // in the target vector.
4495 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4496 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4497 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
4499 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4501 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4503 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4505 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4506 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
4508 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4510 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4512 (X86insrtps VR128:$src1,
4513 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4514 imm:$src3))]>, OpSize;
4517 let Constraints = "$src1 = $dst" in
4518 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4519 let Predicates = [HasAVX] in
4520 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4522 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4523 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4525 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4526 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4527 Requires<[HasSSE41]>;
4529 //===----------------------------------------------------------------------===//
4530 // SSE4.1 - Round Instructions
4531 //===----------------------------------------------------------------------===//
4533 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4534 X86MemOperand x86memop, RegisterClass RC,
4535 PatFrag mem_frag32, PatFrag mem_frag64,
4536 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4537 // Intrinsic operation, reg.
4538 // Vector intrinsic operation, reg
4539 def PSr : SS4AIi8<opcps, MRMSrcReg,
4540 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4541 !strconcat(OpcodeStr,
4542 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4543 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4546 // Vector intrinsic operation, mem
4547 def PSm : Ii8<opcps, MRMSrcMem,
4548 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4549 !strconcat(OpcodeStr,
4550 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4552 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4554 Requires<[HasSSE41]>;
4556 // Vector intrinsic operation, reg
4557 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4558 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4559 !strconcat(OpcodeStr,
4560 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4561 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4564 // Vector intrinsic operation, mem
4565 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4566 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4567 !strconcat(OpcodeStr,
4568 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4570 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4574 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4575 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4576 // Intrinsic operation, reg.
4577 // Vector intrinsic operation, reg
4578 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4579 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4580 !strconcat(OpcodeStr,
4581 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4584 // Vector intrinsic operation, mem
4585 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4586 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4587 !strconcat(OpcodeStr,
4588 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4589 []>, TA, OpSize, Requires<[HasSSE41]>;
4591 // Vector intrinsic operation, reg
4592 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4593 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4594 !strconcat(OpcodeStr,
4595 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4598 // Vector intrinsic operation, mem
4599 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4600 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4601 !strconcat(OpcodeStr,
4602 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4606 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4609 Intrinsic F64Int, bit Is2Addr = 1> {
4610 // Intrinsic operation, reg.
4611 def SSr : SS4AIi8<opcss, MRMSrcReg,
4612 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4614 !strconcat(OpcodeStr,
4615 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4616 !strconcat(OpcodeStr,
4617 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4618 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4621 // Intrinsic operation, mem.
4622 def SSm : SS4AIi8<opcss, MRMSrcMem,
4623 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4625 !strconcat(OpcodeStr,
4626 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4627 !strconcat(OpcodeStr,
4628 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4630 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4633 // Intrinsic operation, reg.
4634 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4635 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4637 !strconcat(OpcodeStr,
4638 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4639 !strconcat(OpcodeStr,
4640 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4641 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4644 // Intrinsic operation, mem.
4645 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4646 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4648 !strconcat(OpcodeStr,
4649 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4650 !strconcat(OpcodeStr,
4651 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4653 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4657 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4659 // Intrinsic operation, reg.
4660 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4661 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4662 !strconcat(OpcodeStr,
4663 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4666 // Intrinsic operation, mem.
4667 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4668 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4669 !strconcat(OpcodeStr,
4670 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4673 // Intrinsic operation, reg.
4674 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4675 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4676 !strconcat(OpcodeStr,
4677 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4680 // Intrinsic operation, mem.
4681 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4682 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4683 !strconcat(OpcodeStr,
4684 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4688 // FP round - roundss, roundps, roundsd, roundpd
4689 let Predicates = [HasAVX] in {
4691 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4692 memopv4f32, memopv2f64,
4693 int_x86_sse41_round_ps,
4694 int_x86_sse41_round_pd>, VEX;
4695 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4696 memopv8f32, memopv4f64,
4697 int_x86_avx_round_ps_256,
4698 int_x86_avx_round_pd_256>, VEX;
4699 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4700 int_x86_sse41_round_ss,
4701 int_x86_sse41_round_sd, 0>, VEX_4V;
4703 // Instructions for the assembler
4704 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4706 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4708 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4711 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4712 memopv4f32, memopv2f64,
4713 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4714 let Constraints = "$src1 = $dst" in
4715 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4716 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4718 //===----------------------------------------------------------------------===//
4719 // SSE4.1 - Packed Bit Test
4720 //===----------------------------------------------------------------------===//
4722 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4723 // the intel intrinsic that corresponds to this.
4724 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4725 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4726 "vptest\t{$src2, $src1|$src1, $src2}",
4727 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4729 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4730 "vptest\t{$src2, $src1|$src1, $src2}",
4731 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4734 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4735 "vptest\t{$src2, $src1|$src1, $src2}",
4736 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4738 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4739 "vptest\t{$src2, $src1|$src1, $src2}",
4740 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4744 let Defs = [EFLAGS] in {
4745 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4746 "ptest \t{$src2, $src1|$src1, $src2}",
4747 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4749 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4750 "ptest \t{$src2, $src1|$src1, $src2}",
4751 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4755 // The bit test instructions below are AVX only
4756 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4757 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4758 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4759 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4760 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4761 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4762 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4763 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4767 let Defs = [EFLAGS], Predicates = [HasAVX] in {
4768 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4769 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4770 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4771 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4774 //===----------------------------------------------------------------------===//
4775 // SSE4.1 - Misc Instructions
4776 //===----------------------------------------------------------------------===//
4778 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4779 "popcnt{w}\t{$src, $dst|$dst, $src}",
4780 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4781 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4782 "popcnt{w}\t{$src, $dst|$dst, $src}",
4783 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4785 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4786 "popcnt{l}\t{$src, $dst|$dst, $src}",
4787 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4788 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4789 "popcnt{l}\t{$src, $dst|$dst, $src}",
4790 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4792 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4793 "popcnt{q}\t{$src, $dst|$dst, $src}",
4794 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4795 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4796 "popcnt{q}\t{$src, $dst|$dst, $src}",
4797 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4801 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4802 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4803 Intrinsic IntId128> {
4804 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4806 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4807 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4808 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4810 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4813 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4816 let Predicates = [HasAVX] in
4817 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4818 int_x86_sse41_phminposuw>, VEX;
4819 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4820 int_x86_sse41_phminposuw>;
4822 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4823 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4824 Intrinsic IntId128, bit Is2Addr = 1> {
4825 let isCommutable = 1 in
4826 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4827 (ins VR128:$src1, VR128:$src2),
4829 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4830 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4831 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4832 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4833 (ins VR128:$src1, i128mem:$src2),
4835 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4836 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4838 (IntId128 VR128:$src1,
4839 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4842 let Predicates = [HasAVX] in {
4843 let isCommutable = 0 in
4844 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4846 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4848 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4850 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4852 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4854 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4856 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4858 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4860 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4862 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4864 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4868 let Constraints = "$src1 = $dst" in {
4869 let isCommutable = 0 in
4870 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4871 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4872 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4873 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4874 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4875 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4876 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4877 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4878 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4879 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4880 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4883 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4884 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4885 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4886 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4888 /// SS48I_binop_rm - Simple SSE41 binary operator.
4889 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4890 ValueType OpVT, bit Is2Addr = 1> {
4891 let isCommutable = 1 in
4892 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4893 (ins VR128:$src1, VR128:$src2),
4895 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4896 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4897 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4899 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4900 (ins VR128:$src1, i128mem:$src2),
4902 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4904 [(set VR128:$dst, (OpNode VR128:$src1,
4905 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4909 let Predicates = [HasAVX] in
4910 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4911 let Constraints = "$src1 = $dst" in
4912 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4914 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4915 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4916 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4917 X86MemOperand x86memop, bit Is2Addr = 1> {
4918 let isCommutable = 1 in
4919 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4920 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
4922 !strconcat(OpcodeStr,
4923 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4924 !strconcat(OpcodeStr,
4925 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4926 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4928 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4929 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
4931 !strconcat(OpcodeStr,
4932 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4933 !strconcat(OpcodeStr,
4934 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4937 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4941 let Predicates = [HasAVX] in {
4942 let isCommutable = 0 in {
4943 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4944 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4945 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4946 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4947 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4948 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4949 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4950 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4951 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4952 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4953 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4954 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4956 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4957 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4958 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4959 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4960 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4961 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4964 let Constraints = "$src1 = $dst" in {
4965 let isCommutable = 0 in {
4966 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4967 VR128, memopv16i8, i128mem>;
4968 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4969 VR128, memopv16i8, i128mem>;
4970 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4971 VR128, memopv16i8, i128mem>;
4972 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4973 VR128, memopv16i8, i128mem>;
4975 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4976 VR128, memopv16i8, i128mem>;
4977 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4978 VR128, memopv16i8, i128mem>;
4981 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4982 let Predicates = [HasAVX] in {
4983 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4984 RegisterClass RC, X86MemOperand x86memop,
4985 PatFrag mem_frag, Intrinsic IntId> {
4986 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4987 (ins RC:$src1, RC:$src2, RC:$src3),
4988 !strconcat(OpcodeStr,
4989 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4990 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4991 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4993 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4994 (ins RC:$src1, x86memop:$src2, RC:$src3),
4995 !strconcat(OpcodeStr,
4996 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4998 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5000 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5004 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5005 memopv16i8, int_x86_sse41_blendvpd>;
5006 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5007 memopv16i8, int_x86_sse41_blendvps>;
5008 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5009 memopv16i8, int_x86_sse41_pblendvb>;
5010 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5011 memopv32i8, int_x86_avx_blendv_pd_256>;
5012 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5013 memopv32i8, int_x86_avx_blendv_ps_256>;
5015 /// SS41I_ternary_int - SSE 4.1 ternary operator
5016 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
5017 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5018 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5019 (ins VR128:$src1, VR128:$src2),
5020 !strconcat(OpcodeStr,
5021 "\t{$src2, $dst|$dst, $src2}"),
5022 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
5025 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5026 (ins VR128:$src1, i128mem:$src2),
5027 !strconcat(OpcodeStr,
5028 "\t{$src2, $dst|$dst, $src2}"),
5031 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
5035 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
5036 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
5037 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
5039 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
5040 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
5042 let Predicates = [HasAVX] in
5043 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5044 "vmovntdqa\t{$src, $dst|$dst, $src}",
5045 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5047 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5048 "movntdqa\t{$src, $dst|$dst, $src}",
5049 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5052 //===----------------------------------------------------------------------===//
5053 // SSE4.2 - Compare Instructions
5054 //===----------------------------------------------------------------------===//
5056 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
5057 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
5058 Intrinsic IntId128, bit Is2Addr = 1> {
5059 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
5060 (ins VR128:$src1, VR128:$src2),
5062 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5063 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5064 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5066 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
5067 (ins VR128:$src1, i128mem:$src2),
5069 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5070 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5072 (IntId128 VR128:$src1,
5073 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5076 let Predicates = [HasAVX] in
5077 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
5079 let Constraints = "$src1 = $dst" in
5080 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
5082 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5083 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
5084 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5085 (PCMPGTQrm VR128:$src1, addr:$src2)>;
5087 //===----------------------------------------------------------------------===//
5088 // SSE4.2 - String/text Processing Instructions
5089 //===----------------------------------------------------------------------===//
5091 // Packed Compare Implicit Length Strings, Return Mask
5092 multiclass pseudo_pcmpistrm<string asm> {
5093 def REG : PseudoI<(outs VR128:$dst),
5094 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5095 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
5097 def MEM : PseudoI<(outs VR128:$dst),
5098 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5099 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
5100 VR128:$src1, (load addr:$src2), imm:$src3))]>;
5103 let Defs = [EFLAGS], usesCustomInserter = 1 in {
5104 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
5105 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
5108 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
5109 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5110 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5111 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5112 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5113 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5114 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5117 let Defs = [XMM0, EFLAGS] in {
5118 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5119 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5120 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5121 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5122 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5123 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5126 // Packed Compare Explicit Length Strings, Return Mask
5127 multiclass pseudo_pcmpestrm<string asm> {
5128 def REG : PseudoI<(outs VR128:$dst),
5129 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5130 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5131 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5132 def MEM : PseudoI<(outs VR128:$dst),
5133 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5134 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5135 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5138 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5139 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5140 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5143 let Predicates = [HasAVX],
5144 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5145 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5146 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5147 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5148 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5149 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5150 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5153 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5154 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5155 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5156 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5157 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5158 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5159 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5162 // Packed Compare Implicit Length Strings, Return Index
5163 let Defs = [ECX, EFLAGS] in {
5164 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5165 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5166 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5167 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5168 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5169 (implicit EFLAGS)]>, OpSize;
5170 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5171 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5172 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5173 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5174 (implicit EFLAGS)]>, OpSize;
5178 let Predicates = [HasAVX] in {
5179 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5181 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5183 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5185 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5187 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5189 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5193 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5194 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5195 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5196 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5197 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5198 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5200 // Packed Compare Explicit Length Strings, Return Index
5201 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5202 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5203 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5204 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5205 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5206 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5207 (implicit EFLAGS)]>, OpSize;
5208 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5209 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5210 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5212 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5213 (implicit EFLAGS)]>, OpSize;
5217 let Predicates = [HasAVX] in {
5218 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5220 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5222 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5224 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5226 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5228 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5232 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5233 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5234 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5235 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5236 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5237 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5239 //===----------------------------------------------------------------------===//
5240 // SSE4.2 - CRC Instructions
5241 //===----------------------------------------------------------------------===//
5243 // No CRC instructions have AVX equivalents
5245 // crc intrinsic instruction
5246 // This set of instructions are only rm, the only difference is the size
5248 let Constraints = "$src1 = $dst" in {
5249 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5250 (ins GR32:$src1, i8mem:$src2),
5251 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5253 (int_x86_sse42_crc32_32_8 GR32:$src1,
5254 (load addr:$src2)))]>;
5255 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5256 (ins GR32:$src1, GR8:$src2),
5257 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5259 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5260 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5261 (ins GR32:$src1, i16mem:$src2),
5262 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5264 (int_x86_sse42_crc32_32_16 GR32:$src1,
5265 (load addr:$src2)))]>,
5267 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5268 (ins GR32:$src1, GR16:$src2),
5269 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5271 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5273 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5274 (ins GR32:$src1, i32mem:$src2),
5275 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5277 (int_x86_sse42_crc32_32_32 GR32:$src1,
5278 (load addr:$src2)))]>;
5279 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5280 (ins GR32:$src1, GR32:$src2),
5281 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5283 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5284 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5285 (ins GR64:$src1, i8mem:$src2),
5286 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5288 (int_x86_sse42_crc32_64_8 GR64:$src1,
5289 (load addr:$src2)))]>,
5291 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5292 (ins GR64:$src1, GR8:$src2),
5293 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5295 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5297 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5298 (ins GR64:$src1, i64mem:$src2),
5299 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5301 (int_x86_sse42_crc32_64_64 GR64:$src1,
5302 (load addr:$src2)))]>,
5304 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5305 (ins GR64:$src1, GR64:$src2),
5306 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5308 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5312 //===----------------------------------------------------------------------===//
5313 // AES-NI Instructions
5314 //===----------------------------------------------------------------------===//
5316 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5317 Intrinsic IntId128, bit Is2Addr = 1> {
5318 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5319 (ins VR128:$src1, VR128:$src2),
5321 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5322 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5323 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5325 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5326 (ins VR128:$src1, i128mem:$src2),
5328 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5331 (IntId128 VR128:$src1,
5332 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5335 // Perform One Round of an AES Encryption/Decryption Flow
5336 let Predicates = [HasAVX, HasAES] in {
5337 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5338 int_x86_aesni_aesenc, 0>, VEX_4V;
5339 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5340 int_x86_aesni_aesenclast, 0>, VEX_4V;
5341 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5342 int_x86_aesni_aesdec, 0>, VEX_4V;
5343 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5344 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5347 let Constraints = "$src1 = $dst" in {
5348 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5349 int_x86_aesni_aesenc>;
5350 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5351 int_x86_aesni_aesenclast>;
5352 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5353 int_x86_aesni_aesdec>;
5354 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5355 int_x86_aesni_aesdeclast>;
5358 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5359 (AESENCrr VR128:$src1, VR128:$src2)>;
5360 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5361 (AESENCrm VR128:$src1, addr:$src2)>;
5362 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5363 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5364 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5365 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5366 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5367 (AESDECrr VR128:$src1, VR128:$src2)>;
5368 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5369 (AESDECrm VR128:$src1, addr:$src2)>;
5370 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5371 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5372 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5373 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5375 // Perform the AES InvMixColumn Transformation
5376 let Predicates = [HasAVX, HasAES] in {
5377 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5379 "vaesimc\t{$src1, $dst|$dst, $src1}",
5381 (int_x86_aesni_aesimc VR128:$src1))]>,
5383 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5384 (ins i128mem:$src1),
5385 "vaesimc\t{$src1, $dst|$dst, $src1}",
5387 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5390 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5392 "aesimc\t{$src1, $dst|$dst, $src1}",
5394 (int_x86_aesni_aesimc VR128:$src1))]>,
5396 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5397 (ins i128mem:$src1),
5398 "aesimc\t{$src1, $dst|$dst, $src1}",
5400 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5403 // AES Round Key Generation Assist
5404 let Predicates = [HasAVX, HasAES] in {
5405 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5406 (ins VR128:$src1, i8imm:$src2),
5407 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5409 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5411 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5412 (ins i128mem:$src1, i8imm:$src2),
5413 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5415 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5419 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5420 (ins VR128:$src1, i8imm:$src2),
5421 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5423 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5425 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5426 (ins i128mem:$src1, i8imm:$src2),
5427 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5429 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5433 //===----------------------------------------------------------------------===//
5434 // CLMUL Instructions
5435 //===----------------------------------------------------------------------===//
5437 // Carry-less Multiplication instructions
5438 let Constraints = "$src1 = $dst" in {
5439 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5440 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5441 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5444 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5445 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5446 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5450 // AVX carry-less Multiplication instructions
5451 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5452 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5453 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5456 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5457 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5458 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5462 multiclass pclmul_alias<string asm, int immop> {
5463 def : InstAlias<!strconcat("pclmul", asm,
5464 "dq {$src, $dst|$dst, $src}"),
5465 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5467 def : InstAlias<!strconcat("pclmul", asm,
5468 "dq {$src, $dst|$dst, $src}"),
5469 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5471 def : InstAlias<!strconcat("vpclmul", asm,
5472 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5473 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5475 def : InstAlias<!strconcat("vpclmul", asm,
5476 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5477 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5479 defm : pclmul_alias<"hqhq", 0x11>;
5480 defm : pclmul_alias<"hqlq", 0x01>;
5481 defm : pclmul_alias<"lqhq", 0x10>;
5482 defm : pclmul_alias<"lqlq", 0x00>;
5484 //===----------------------------------------------------------------------===//
5486 //===----------------------------------------------------------------------===//
5488 //===----------------------------------------------------------------------===//
5489 // VBROADCAST - Load from memory and broadcast to all elements of the
5490 // destination operand
5492 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5493 X86MemOperand x86memop, Intrinsic Int> :
5494 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5495 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5496 [(set RC:$dst, (Int addr:$src))]>, VEX;
5498 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5499 int_x86_avx_vbroadcastss>;
5500 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5501 int_x86_avx_vbroadcastss_256>;
5502 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5503 int_x86_avx_vbroadcast_sd_256>;
5504 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5505 int_x86_avx_vbroadcastf128_pd_256>;
5507 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5508 (VBROADCASTF128 addr:$src)>;
5510 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
5511 (VBROADCASTSSY addr:$src)>;
5512 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
5513 (VBROADCASTSD addr:$src)>;
5514 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
5515 (VBROADCASTSSY addr:$src)>;
5516 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
5517 (VBROADCASTSD addr:$src)>;
5519 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
5520 (VBROADCASTSS addr:$src)>;
5521 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
5522 (VBROADCASTSS addr:$src)>;
5524 //===----------------------------------------------------------------------===//
5525 // VINSERTF128 - Insert packed floating-point values
5527 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5528 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5529 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5531 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5532 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5533 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5536 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5537 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5538 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5539 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5540 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5541 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5543 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5545 (VINSERTF128rr VR256:$src1, VR128:$src2,
5546 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5547 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5549 (VINSERTF128rr VR256:$src1, VR128:$src2,
5550 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5551 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5553 (VINSERTF128rr VR256:$src1, VR128:$src2,
5554 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5555 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5557 (VINSERTF128rr VR256:$src1, VR128:$src2,
5558 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5559 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
5561 (VINSERTF128rr VR256:$src1, VR128:$src2,
5562 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5563 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
5565 (VINSERTF128rr VR256:$src1, VR128:$src2,
5566 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5568 // Special COPY patterns
5569 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
5570 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5571 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
5572 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5573 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
5574 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5575 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
5576 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5577 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
5578 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5579 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
5580 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5582 //===----------------------------------------------------------------------===//
5583 // VEXTRACTF128 - Extract packed floating-point values
5585 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5586 (ins VR256:$src1, i8imm:$src2),
5587 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5589 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5590 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5591 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5594 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5595 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5596 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5597 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5598 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5599 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5601 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5602 (v4f32 (VEXTRACTF128rr
5603 (v8f32 VR256:$src1),
5604 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5605 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5606 (v2f64 (VEXTRACTF128rr
5607 (v4f64 VR256:$src1),
5608 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5609 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5610 (v4i32 (VEXTRACTF128rr
5611 (v8i32 VR256:$src1),
5612 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5613 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5614 (v2i64 (VEXTRACTF128rr
5615 (v4i64 VR256:$src1),
5616 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5617 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5618 (v8i16 (VEXTRACTF128rr
5619 (v16i16 VR256:$src1),
5620 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5621 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5622 (v16i8 (VEXTRACTF128rr
5623 (v32i8 VR256:$src1),
5624 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5626 // Special COPY patterns
5627 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
5628 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
5629 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
5630 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
5632 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
5633 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
5634 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
5635 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
5638 //===----------------------------------------------------------------------===//
5639 // VMASKMOV - Conditional SIMD Packed Loads and Stores
5641 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5642 Intrinsic IntLd, Intrinsic IntLd256,
5643 Intrinsic IntSt, Intrinsic IntSt256,
5644 PatFrag pf128, PatFrag pf256> {
5645 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5646 (ins VR128:$src1, f128mem:$src2),
5647 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5648 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5650 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5651 (ins VR256:$src1, f256mem:$src2),
5652 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5653 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5655 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5656 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5657 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5658 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5659 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5660 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5661 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5662 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5665 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5666 int_x86_avx_maskload_ps,
5667 int_x86_avx_maskload_ps_256,
5668 int_x86_avx_maskstore_ps,
5669 int_x86_avx_maskstore_ps_256,
5670 memopv4f32, memopv8f32>;
5671 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5672 int_x86_avx_maskload_pd,
5673 int_x86_avx_maskload_pd_256,
5674 int_x86_avx_maskstore_pd,
5675 int_x86_avx_maskstore_pd_256,
5676 memopv2f64, memopv4f64>;
5678 //===----------------------------------------------------------------------===//
5679 // VPERMIL - Permute Single and Double Floating-Point Values
5681 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5682 RegisterClass RC, X86MemOperand x86memop_f,
5683 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5684 Intrinsic IntVar, Intrinsic IntImm> {
5685 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5686 (ins RC:$src1, RC:$src2),
5687 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5688 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5689 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5690 (ins RC:$src1, x86memop_i:$src2),
5691 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5692 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5694 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5695 (ins RC:$src1, i8imm:$src2),
5696 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5697 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5698 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5699 (ins x86memop_f:$src1, i8imm:$src2),
5700 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5701 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5704 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5705 memopv4f32, memopv4i32,
5706 int_x86_avx_vpermilvar_ps,
5707 int_x86_avx_vpermil_ps>;
5708 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5709 memopv8f32, memopv8i32,
5710 int_x86_avx_vpermilvar_ps_256,
5711 int_x86_avx_vpermil_ps_256>;
5712 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5713 memopv2f64, memopv2i64,
5714 int_x86_avx_vpermilvar_pd,
5715 int_x86_avx_vpermil_pd>;
5716 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5717 memopv4f64, memopv4i64,
5718 int_x86_avx_vpermilvar_pd_256,
5719 int_x86_avx_vpermil_pd_256>;
5721 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
5722 (VPERMILPSYri VR256:$src1, imm:$imm)>;
5723 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
5724 (VPERMILPDYri VR256:$src1, imm:$imm)>;
5725 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
5726 (VPERMILPSYri VR256:$src1, imm:$imm)>;
5727 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
5728 (VPERMILPDYri VR256:$src1, imm:$imm)>;
5730 //===----------------------------------------------------------------------===//
5731 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
5733 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5734 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5735 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5737 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5738 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5739 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5742 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5743 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5744 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5745 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5746 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5747 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5749 def : Pat<(int_x86_avx_vperm2f128_ps_256
5750 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5751 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5752 def : Pat<(int_x86_avx_vperm2f128_pd_256
5753 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5754 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5755 def : Pat<(int_x86_avx_vperm2f128_si_256
5756 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5757 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5759 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5760 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5761 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5762 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5763 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5764 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5765 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5766 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5767 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5768 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5769 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
5770 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
5772 //===----------------------------------------------------------------------===//
5773 // VZERO - Zero YMM registers
5775 // Zero All YMM registers
5776 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5777 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5779 // Zero Upper bits of YMM registers
5780 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5781 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5783 //===----------------------------------------------------------------------===//
5784 // SSE Shuffle pattern fragments
5785 //===----------------------------------------------------------------------===//
5787 // This is part of a "work in progress" refactoring. The idea is that all
5788 // vector shuffles are going to be translated into target specific nodes and
5789 // directly matched by the patterns below (which can be changed along the way)
5790 // The AVX version of some but not all of them are described here, and more
5791 // should come in a near future.
5793 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5794 // SSE2 loads, which are always promoted to v2i64. The last one should match
5795 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5796 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5797 // we investigate further.
5798 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5800 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5801 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5803 (PSHUFDmi addr:$src1, imm:$imm)>;
5804 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5806 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5808 // Shuffle with PSHUFD instruction.
5809 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5810 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5811 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5812 (PSHUFDri VR128:$src1, imm:$imm)>;
5814 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5815 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5816 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5817 (PSHUFDri VR128:$src1, imm:$imm)>;
5819 // Shuffle with SHUFPD instruction.
5820 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5821 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5822 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5823 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5824 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5825 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5827 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5828 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5829 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5830 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5832 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5833 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5834 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5835 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5837 // Shuffle with SHUFPS instruction.
5838 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5839 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5840 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5841 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5842 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5843 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5845 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5846 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5847 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5848 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5850 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5851 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5852 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5853 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5854 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5855 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5857 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5858 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5859 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5860 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5862 // Shuffle with MOVHLPS instruction
5863 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5864 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5865 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5866 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5868 // Shuffle with MOVDDUP instruction
5869 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5870 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5871 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5872 (MOVDDUPrm addr:$src)>;
5874 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5875 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5876 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5877 (MOVDDUPrm addr:$src)>;
5879 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5880 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5881 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5882 (MOVDDUPrm addr:$src)>;
5884 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5885 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5886 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5887 (MOVDDUPrm addr:$src)>;
5889 def : Pat<(X86Movddup (bc_v2f64
5890 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5891 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5892 def : Pat<(X86Movddup (bc_v2f64
5893 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5894 (MOVDDUPrm addr:$src)>;
5897 // Shuffle with UNPCKLPS
5898 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5899 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5900 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5901 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5903 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5904 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5905 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5906 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5908 // Shuffle with VUNPCKHPSY
5909 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
5910 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5911 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5912 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5913 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
5914 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5915 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
5916 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5918 // Shuffle with UNPCKHPS
5919 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5920 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5921 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5922 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5924 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5925 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5926 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5927 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5929 // Shuffle with VUNPCKHPSY
5930 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
5931 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5932 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
5933 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5935 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
5936 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5937 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
5938 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5940 // Shuffle with UNPCKLPD
5941 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5942 (VUNPCKLPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5943 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5944 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
5946 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5947 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5948 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5949 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5951 // Shuffle with VUNPCKLPDY
5952 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
5953 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5954 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5955 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5957 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
5958 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5959 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
5960 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5962 // Shuffle with UNPCKHPD
5963 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5964 (VUNPCKHPDrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5965 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5966 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
5968 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5969 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5970 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5971 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5973 // Shuffle with VUNPCKHPDY
5974 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
5975 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5976 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
5977 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5978 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
5979 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>, Requires<[HasAVX]>;
5980 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
5981 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>, Requires<[HasAVX]>;
5983 // Shuffle with MOVLHPS
5984 def : Pat<(X86Movlhps VR128:$src1,
5985 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5986 (MOVHPSrm VR128:$src1, addr:$src2)>;
5987 def : Pat<(X86Movlhps VR128:$src1,
5988 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5989 (MOVHPSrm VR128:$src1, addr:$src2)>;
5990 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5991 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5992 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5993 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5994 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5995 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5997 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5998 // is during lowering, where it's not possible to recognize the load fold cause
5999 // it has two uses through a bitcast. One use disappears at isel time and the
6000 // fold opportunity reappears.
6001 def : Pat<(v2f64 (X86Movddup VR128:$src)),
6002 (UNPCKLPDrr VR128:$src, VR128:$src)>;
6004 // Shuffle with MOVLHPD
6005 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
6006 (scalar_to_vector (loadf64 addr:$src2)))),
6007 (MOVHPDrm VR128:$src1, addr:$src2)>;
6009 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
6010 // is during lowering, where it's not possible to recognize the load fold cause
6011 // it has two uses through a bitcast. One use disappears at isel time and the
6012 // fold opportunity reappears.
6013 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
6014 (scalar_to_vector (loadf64 addr:$src2)))),
6015 (MOVHPDrm VR128:$src1, addr:$src2)>;
6017 // Shuffle with MOVSS
6018 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
6019 (MOVSSrr VR128:$src1, FR32:$src2)>;
6020 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
6021 (MOVSSrr (v4i32 VR128:$src1),
6022 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
6023 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
6024 (MOVSSrr (v4f32 VR128:$src1),
6025 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
6027 // Shuffle with MOVSD
6028 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
6029 (MOVSDrr VR128:$src1, FR64:$src2)>;
6030 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
6031 (MOVSDrr (v2i64 VR128:$src1),
6032 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
6033 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
6034 (MOVSDrr (v2f64 VR128:$src1),
6035 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
6036 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
6037 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6038 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
6039 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6041 // Shuffle with PSHUFHW
6042 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
6043 (PSHUFHWri VR128:$src, imm:$imm)>;
6044 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
6045 (PSHUFHWmi addr:$src, imm:$imm)>;
6047 // Shuffle with PSHUFLW
6048 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
6049 (PSHUFLWri VR128:$src, imm:$imm)>;
6050 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
6051 (PSHUFLWmi addr:$src, imm:$imm)>;
6053 // Shuffle with MOVLPS
6054 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
6055 (MOVLPSrm VR128:$src1, addr:$src2)>;
6056 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
6057 (MOVLPSrm VR128:$src1, addr:$src2)>;
6058 def : Pat<(X86Movlps VR128:$src1,
6059 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
6060 (MOVLPSrm VR128:$src1, addr:$src2)>;
6061 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
6062 // is during lowering, where it's not possible to recognize the load fold cause
6063 // it has two uses through a bitcast. One use disappears at isel time and the
6064 // fold opportunity reappears.
6065 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
6066 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6068 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
6069 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6071 // Shuffle with MOVLPD
6072 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6073 (MOVLPDrm VR128:$src1, addr:$src2)>;
6074 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6075 (MOVLPDrm VR128:$src1, addr:$src2)>;
6076 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
6077 (scalar_to_vector (loadf64 addr:$src2)))),
6078 (MOVLPDrm VR128:$src1, addr:$src2)>;
6080 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
6081 def : Pat<(store (f64 (vector_extract
6082 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6083 (MOVHPSmr addr:$dst, VR128:$src)>;
6084 def : Pat<(store (f64 (vector_extract
6085 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6086 (MOVHPDmr addr:$dst, VR128:$src)>;
6088 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
6089 (MOVLPSmr addr:$src1, VR128:$src2)>;
6090 def : Pat<(store (v4i32 (X86Movlps
6091 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
6092 (MOVLPSmr addr:$src1, VR128:$src2)>;
6094 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6095 (MOVLPDmr addr:$src1, VR128:$src2)>;
6096 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6097 (MOVLPDmr addr:$src1, VR128:$src2)>;