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:$src2)))]>;
476 let Predicates = [HasAVX] in {
478 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
479 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
480 def : Pat<(X86Movlhps VR128:$src1,
481 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
482 (VMOVHPSrm VR128:$src1, addr:$src2)>;
483 def : Pat<(X86Movlhps VR128:$src1,
484 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
485 (VMOVHPSrm VR128:$src1, addr:$src2)>;
488 let AddedComplexity = 20 in {
489 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
490 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
491 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
492 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
494 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
495 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
496 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
498 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
499 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
500 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
501 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
502 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
503 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
506 let AddedComplexity = 20 in {
507 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
508 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
509 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
511 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
512 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
513 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
514 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
515 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
518 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
519 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
520 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
521 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
524 let Predicates = [HasSSE1] in {
526 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
527 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
528 def : Pat<(X86Movlhps VR128:$src1,
529 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
530 (MOVHPSrm VR128:$src1, addr:$src2)>;
531 def : Pat<(X86Movlhps VR128:$src1,
532 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
533 (MOVHPSrm VR128:$src1, addr:$src2)>;
536 let AddedComplexity = 20 in {
537 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
538 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
539 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
540 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
542 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
543 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
544 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
546 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
547 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
548 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
549 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
550 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
551 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
554 let AddedComplexity = 20 in {
555 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
556 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
557 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
559 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
560 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
561 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
562 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
563 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
566 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
567 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
568 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
569 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
572 //===----------------------------------------------------------------------===//
573 // SSE 1 & 2 - Conversion Instructions
574 //===----------------------------------------------------------------------===//
576 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
577 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
579 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
580 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
581 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
582 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
585 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
586 X86MemOperand x86memop, string asm> {
587 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
589 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
593 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
594 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
595 string asm, Domain d> {
596 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
597 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
598 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
599 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
602 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
603 X86MemOperand x86memop, string asm> {
604 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
605 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
606 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
607 (ins DstRC:$src1, x86memop:$src),
608 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
611 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
612 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
613 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
614 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
616 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
617 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
618 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
619 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
622 // The assembler can recognize rr 64-bit instructions by seeing a rxx
623 // register, but the same isn't true when only using memory operands,
624 // provide other assembly "l" and "q" forms to address this explicitly
625 // where appropriate to do so.
626 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
628 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
630 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
632 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
634 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
637 let Predicates = [HasAVX] in {
638 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
639 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
640 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
641 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
642 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
643 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
644 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
645 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
647 def : Pat<(f32 (sint_to_fp GR32:$src)),
648 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
649 def : Pat<(f32 (sint_to_fp GR64:$src)),
650 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
651 def : Pat<(f64 (sint_to_fp GR32:$src)),
652 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
653 def : Pat<(f64 (sint_to_fp GR64:$src)),
654 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
657 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
658 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
659 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
660 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
661 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
662 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
663 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
664 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
665 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
666 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
667 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
668 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
669 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
670 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
671 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
672 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
674 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
675 // and/or XMM operand(s).
677 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
678 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
680 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
681 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
682 [(set DstRC:$dst, (Int SrcRC:$src))]>;
683 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
684 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
685 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
688 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
689 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
690 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
691 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
693 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
694 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
695 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
696 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
697 (ins DstRC:$src1, x86memop:$src2),
699 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
700 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
701 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
704 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
705 f128mem, load, "cvtsd2si">, XD, VEX;
706 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
707 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
710 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
711 // Get rid of this hack or rename the intrinsics, there are several
712 // intructions that only match with the intrinsic form, why create duplicates
713 // to let them be recognized by the assembler?
714 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
715 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
716 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
717 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
718 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
719 f128mem, load, "cvtsd2si{l}">, XD;
720 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
721 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
724 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
725 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
726 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
727 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
729 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
730 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
731 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
732 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
735 let Constraints = "$src1 = $dst" in {
736 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
737 int_x86_sse_cvtsi2ss, i32mem, loadi32,
739 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
740 int_x86_sse_cvtsi642ss, i64mem, loadi64,
741 "cvtsi2ss{q}">, XS, REX_W;
742 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
743 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
745 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
746 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
747 "cvtsi2sd">, XD, REX_W;
752 // Aliases for intrinsics
753 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
754 f32mem, load, "cvttss2si">, XS, VEX;
755 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
756 int_x86_sse_cvttss2si64, f32mem, load,
757 "cvttss2si">, XS, VEX, VEX_W;
758 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
759 f128mem, load, "cvttsd2si">, XD, VEX;
760 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
761 int_x86_sse2_cvttsd2si64, f128mem, load,
762 "cvttsd2si">, XD, VEX, VEX_W;
763 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
764 f32mem, load, "cvttss2si">, XS;
765 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
766 int_x86_sse_cvttss2si64, f32mem, load,
767 "cvttss2si{q}">, XS, REX_W;
768 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
769 f128mem, load, "cvttsd2si">, XD;
770 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
771 int_x86_sse2_cvttsd2si64, f128mem, load,
772 "cvttsd2si{q}">, XD, REX_W;
774 let Pattern = []<dag> in {
775 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
776 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
777 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
778 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
780 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
781 "cvtdq2ps\t{$src, $dst|$dst, $src}",
782 SSEPackedSingle>, TB, VEX;
783 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
784 "cvtdq2ps\t{$src, $dst|$dst, $src}",
785 SSEPackedSingle>, TB, VEX;
788 let Pattern = []<dag> in {
789 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
790 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
791 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
792 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
793 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
794 "cvtdq2ps\t{$src, $dst|$dst, $src}",
795 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
798 let Predicates = [HasSSE1] in {
799 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
800 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
801 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
802 (CVTSS2SIrm addr:$src)>;
803 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
804 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
805 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
806 (CVTSS2SI64rm addr:$src)>;
809 let Predicates = [HasAVX] in {
810 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
811 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
812 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
813 (VCVTSS2SIrm addr:$src)>;
814 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
815 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
816 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
817 (VCVTSS2SI64rm addr:$src)>;
822 // Convert scalar double to scalar single
823 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
824 (ins FR64:$src1, FR64:$src2),
825 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
827 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
828 (ins FR64:$src1, f64mem:$src2),
829 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
830 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
831 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
834 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
835 "cvtsd2ss\t{$src, $dst|$dst, $src}",
836 [(set FR32:$dst, (fround FR64:$src))]>;
837 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
838 "cvtsd2ss\t{$src, $dst|$dst, $src}",
839 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
840 Requires<[HasSSE2, OptForSize]>;
842 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
843 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
845 let Constraints = "$src1 = $dst" in
846 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
847 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
849 // Convert scalar single to scalar double
850 // SSE2 instructions with XS prefix
851 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
852 (ins FR32:$src1, FR32:$src2),
853 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
854 []>, XS, Requires<[HasAVX]>, VEX_4V;
855 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
856 (ins FR32:$src1, f32mem:$src2),
857 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
858 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
860 let Predicates = [HasAVX] in {
861 def : Pat<(f64 (fextend FR32:$src)),
862 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
863 def : Pat<(fextend (loadf32 addr:$src)),
864 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
865 def : Pat<(extloadf32 addr:$src),
866 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
869 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
870 "cvtss2sd\t{$src, $dst|$dst, $src}",
871 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
873 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
874 "cvtss2sd\t{$src, $dst|$dst, $src}",
875 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
876 Requires<[HasSSE2, OptForSize]>;
878 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
879 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
880 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
881 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
882 VR128:$src2))]>, XS, VEX_4V,
884 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
885 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
886 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
887 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
888 (load addr:$src2)))]>, XS, VEX_4V,
890 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
891 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
892 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
893 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
894 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
897 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
898 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
899 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
900 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
901 (load addr:$src2)))]>, XS,
905 def : Pat<(extloadf32 addr:$src),
906 (CVTSS2SDrr (MOVSSrm addr:$src))>,
907 Requires<[HasSSE2, OptForSpeed]>;
909 // Convert doubleword to packed single/double fp
910 // SSE2 instructions without OpSize prefix
911 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
912 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
913 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
914 TB, VEX, Requires<[HasAVX]>;
915 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
916 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
917 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
918 (bitconvert (memopv2i64 addr:$src))))]>,
919 TB, VEX, Requires<[HasAVX]>;
920 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
921 "cvtdq2ps\t{$src, $dst|$dst, $src}",
922 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
923 TB, Requires<[HasSSE2]>;
924 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
925 "cvtdq2ps\t{$src, $dst|$dst, $src}",
926 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
927 (bitconvert (memopv2i64 addr:$src))))]>,
928 TB, Requires<[HasSSE2]>;
930 // FIXME: why the non-intrinsic version is described as SSE3?
931 // SSE2 instructions with XS prefix
932 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
933 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
934 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
935 XS, VEX, Requires<[HasAVX]>;
936 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
937 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
938 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
939 (bitconvert (memopv2i64 addr:$src))))]>,
940 XS, VEX, Requires<[HasAVX]>;
941 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
942 "cvtdq2pd\t{$src, $dst|$dst, $src}",
943 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
944 XS, Requires<[HasSSE2]>;
945 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
946 "cvtdq2pd\t{$src, $dst|$dst, $src}",
947 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
948 (bitconvert (memopv2i64 addr:$src))))]>,
949 XS, Requires<[HasSSE2]>;
952 // Convert packed single/double fp to doubleword
953 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
954 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
955 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
956 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
957 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
958 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
959 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
960 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
961 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
962 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
963 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
964 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
966 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
967 "cvtps2dq\t{$src, $dst|$dst, $src}",
968 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
970 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
972 "cvtps2dq\t{$src, $dst|$dst, $src}",
973 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
974 (memop addr:$src)))]>, VEX;
975 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
976 "cvtps2dq\t{$src, $dst|$dst, $src}",
977 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
978 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
979 "cvtps2dq\t{$src, $dst|$dst, $src}",
980 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
981 (memop addr:$src)))]>;
983 // SSE2 packed instructions with XD prefix
984 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
985 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
986 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
987 XD, VEX, Requires<[HasAVX]>;
988 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
989 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
990 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
991 (memop addr:$src)))]>,
992 XD, VEX, Requires<[HasAVX]>;
993 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
994 "cvtpd2dq\t{$src, $dst|$dst, $src}",
995 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
996 XD, Requires<[HasSSE2]>;
997 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
998 "cvtpd2dq\t{$src, $dst|$dst, $src}",
999 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1000 (memop addr:$src)))]>,
1001 XD, Requires<[HasSSE2]>;
1004 // Convert with truncation packed single/double fp to doubleword
1005 // SSE2 packed instructions with XS prefix
1006 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1007 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1008 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1009 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1010 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1011 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1012 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1013 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1014 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1015 "cvttps2dq\t{$src, $dst|$dst, $src}",
1017 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1018 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1019 "cvttps2dq\t{$src, $dst|$dst, $src}",
1021 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1023 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1024 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1026 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1027 XS, VEX, Requires<[HasAVX]>;
1028 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1029 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1030 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1031 (memop addr:$src)))]>,
1032 XS, VEX, Requires<[HasAVX]>;
1034 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1035 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
1036 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1037 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
1039 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1040 (Int_VCVTDQ2PSrr VR128:$src)>, Requires<[HasAVX]>;
1041 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1042 (VCVTTPS2DQrr VR128:$src)>, Requires<[HasAVX]>;
1043 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1044 (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
1045 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1046 (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
1048 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
1050 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1051 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1053 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1055 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1056 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1057 (memop addr:$src)))]>, VEX;
1058 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1059 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1060 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1061 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1062 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1063 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1064 (memop addr:$src)))]>;
1066 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1067 // register, but the same isn't true when using memory operands instead.
1068 // Provide other assembly rr and rm forms to address this explicitly.
1069 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1070 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1071 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1072 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1075 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1076 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1077 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1078 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1081 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1082 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1083 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1084 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1086 // Convert packed single to packed double
1087 let Predicates = [HasAVX] in {
1088 // SSE2 instructions without OpSize prefix
1089 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1090 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1091 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1092 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1093 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1094 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1095 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1096 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
1098 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1099 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1100 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1101 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1103 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1104 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1105 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1106 VEX, Requires<[HasAVX]>;
1107 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1108 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1109 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1110 (load addr:$src)))]>,
1111 VEX, Requires<[HasAVX]>;
1112 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1113 "cvtps2pd\t{$src, $dst|$dst, $src}",
1114 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1115 TB, Requires<[HasSSE2]>;
1116 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1117 "cvtps2pd\t{$src, $dst|$dst, $src}",
1118 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1119 (load addr:$src)))]>,
1120 TB, Requires<[HasSSE2]>;
1122 // Convert packed double to packed single
1123 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1124 // register, but the same isn't true when using memory operands instead.
1125 // Provide other assembly rr and rm forms to address this explicitly.
1126 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1127 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1128 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1129 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1132 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1133 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1134 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1135 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1138 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1139 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1140 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1141 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1142 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1143 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1144 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1145 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1148 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1149 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1150 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1151 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1153 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1154 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1155 (memop addr:$src)))]>;
1156 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1157 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1158 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1159 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1160 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1161 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1162 (memop addr:$src)))]>;
1164 // AVX 256-bit register conversion intrinsics
1165 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1166 // whenever possible to avoid declaring two versions of each one.
1167 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1168 (VCVTDQ2PSYrr VR256:$src)>;
1169 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1170 (VCVTDQ2PSYrm addr:$src)>;
1172 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1173 (VCVTPD2PSYrr VR256:$src)>;
1174 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1175 (VCVTPD2PSYrm addr:$src)>;
1177 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1178 (VCVTPS2DQYrr VR256:$src)>;
1179 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1180 (VCVTPS2DQYrm addr:$src)>;
1182 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1183 (VCVTPS2PDYrr VR128:$src)>;
1184 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1185 (VCVTPS2PDYrm addr:$src)>;
1187 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1188 (VCVTTPD2DQYrr VR256:$src)>;
1189 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1190 (VCVTTPD2DQYrm addr:$src)>;
1192 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1193 (VCVTTPS2DQYrr VR256:$src)>;
1194 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1195 (VCVTTPS2DQYrm addr:$src)>;
1197 // Match fround and fextend for 128/256-bit conversions
1198 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1199 (VCVTPD2PSYrr VR256:$src)>;
1200 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1201 (VCVTPD2PSYrm addr:$src)>;
1203 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1204 (VCVTPS2PDYrr VR128:$src)>;
1205 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1206 (VCVTPS2PDYrm addr:$src)>;
1208 //===----------------------------------------------------------------------===//
1209 // SSE 1 & 2 - Compare Instructions
1210 //===----------------------------------------------------------------------===//
1212 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1213 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1214 string asm, string asm_alt> {
1215 let isAsmParserOnly = 1 in {
1216 def rr : SIi8<0xC2, MRMSrcReg,
1217 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1220 def rm : SIi8<0xC2, MRMSrcMem,
1221 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1225 // Accept explicit immediate argument form instead of comparison code.
1226 def rr_alt : SIi8<0xC2, MRMSrcReg,
1227 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1230 def rm_alt : SIi8<0xC2, MRMSrcMem,
1231 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1235 let neverHasSideEffects = 1 in {
1236 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1237 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1238 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1240 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1241 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1242 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1246 let Constraints = "$src1 = $dst" in {
1247 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1248 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1249 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1250 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1251 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1252 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1253 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1254 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1255 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1256 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1257 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1258 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1259 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1260 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1261 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1262 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1264 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1265 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1266 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1267 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1268 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1269 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1270 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1271 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1272 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1273 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1274 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1275 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1276 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1279 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1280 Intrinsic Int, string asm> {
1281 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1282 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1283 [(set VR128:$dst, (Int VR128:$src1,
1284 VR128:$src, imm:$cc))]>;
1285 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1286 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1287 [(set VR128:$dst, (Int VR128:$src1,
1288 (load addr:$src), imm:$cc))]>;
1291 // Aliases to match intrinsics which expect XMM operand(s).
1292 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1293 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1295 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1296 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1298 let Constraints = "$src1 = $dst" in {
1299 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1300 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1301 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1302 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1306 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1307 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1308 ValueType vt, X86MemOperand x86memop,
1309 PatFrag ld_frag, string OpcodeStr, Domain d> {
1310 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1311 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1312 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1313 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1314 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1315 [(set EFLAGS, (OpNode (vt RC:$src1),
1316 (ld_frag addr:$src2)))], d>;
1319 let Defs = [EFLAGS] in {
1320 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1321 "ucomiss", SSEPackedSingle>, VEX;
1322 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1323 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1324 let Pattern = []<dag> in {
1325 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1326 "comiss", SSEPackedSingle>, VEX;
1327 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1328 "comisd", SSEPackedDouble>, OpSize, VEX;
1331 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1332 load, "ucomiss", SSEPackedSingle>, VEX;
1333 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1334 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1336 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1337 load, "comiss", SSEPackedSingle>, VEX;
1338 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1339 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1340 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1341 "ucomiss", SSEPackedSingle>, TB;
1342 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1343 "ucomisd", SSEPackedDouble>, TB, OpSize;
1345 let Pattern = []<dag> in {
1346 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1347 "comiss", SSEPackedSingle>, TB;
1348 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1349 "comisd", SSEPackedDouble>, TB, OpSize;
1352 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1353 load, "ucomiss", SSEPackedSingle>, TB;
1354 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1355 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1357 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1358 "comiss", SSEPackedSingle>, TB;
1359 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1360 "comisd", SSEPackedDouble>, TB, OpSize;
1361 } // Defs = [EFLAGS]
1363 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1364 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1365 Intrinsic Int, string asm, string asm_alt,
1367 let isAsmParserOnly = 1 in {
1368 def rri : PIi8<0xC2, MRMSrcReg,
1369 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1370 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1371 def rmi : PIi8<0xC2, MRMSrcMem,
1372 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1373 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1376 // Accept explicit immediate argument form instead of comparison code.
1377 def rri_alt : PIi8<0xC2, MRMSrcReg,
1378 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1380 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1381 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1385 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1386 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1387 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1388 SSEPackedSingle>, VEX_4V;
1389 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1390 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1391 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1392 SSEPackedDouble>, OpSize, VEX_4V;
1393 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1394 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1395 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1396 SSEPackedSingle>, VEX_4V;
1397 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1398 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1399 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1400 SSEPackedDouble>, OpSize, VEX_4V;
1401 let Constraints = "$src1 = $dst" in {
1402 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1403 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1404 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1405 SSEPackedSingle>, TB;
1406 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1407 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1408 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1409 SSEPackedDouble>, TB, OpSize;
1412 let Predicates = [HasSSE1] in {
1413 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1414 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1415 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1416 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1419 let Predicates = [HasSSE2] in {
1420 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1421 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1422 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1423 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1426 let Predicates = [HasAVX] in {
1427 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1428 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1429 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1430 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1431 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1432 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1433 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1434 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1436 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
1437 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
1438 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
1439 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
1440 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
1441 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
1442 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
1443 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
1446 //===----------------------------------------------------------------------===//
1447 // SSE 1 & 2 - Shuffle Instructions
1448 //===----------------------------------------------------------------------===//
1450 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1451 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1452 ValueType vt, string asm, PatFrag mem_frag,
1453 Domain d, bit IsConvertibleToThreeAddress = 0> {
1454 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1455 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1456 [(set RC:$dst, (vt (shufp:$src3
1457 RC:$src1, (mem_frag addr:$src2))))], d>;
1458 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1459 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1460 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1462 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1465 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1466 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1467 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1468 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1469 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1470 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1471 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1472 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1473 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1474 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1475 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1476 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1478 let Constraints = "$src1 = $dst" in {
1479 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1480 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1481 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1483 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1484 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1485 memopv2f64, SSEPackedDouble>, TB, OpSize;
1488 let Predicates = [HasSSE1] in {
1489 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1490 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1491 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1492 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1493 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1494 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1495 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1496 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1497 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1498 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1499 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1500 // fall back to this for SSE1)
1501 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1502 (SHUFPSrri VR128:$src2, VR128:$src1,
1503 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1504 // Special unary SHUFPSrri case.
1505 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1506 (SHUFPSrri VR128:$src1, VR128:$src1,
1507 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1510 let Predicates = [HasSSE2] in {
1511 // Special binary v4i32 shuffle cases with SHUFPS.
1512 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1513 (SHUFPSrri VR128:$src1, VR128:$src2,
1514 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1515 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1516 (bc_v4i32 (memopv2i64 addr:$src2)))),
1517 (SHUFPSrmi VR128:$src1, addr:$src2,
1518 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1519 // Special unary SHUFPDrri cases.
1520 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1521 (SHUFPDrri VR128:$src1, VR128:$src1,
1522 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1523 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1524 (SHUFPDrri VR128:$src1, VR128:$src1,
1525 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1526 // Special binary v2i64 shuffle cases using SHUFPDrri.
1527 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1528 (SHUFPDrri VR128:$src1, VR128:$src2,
1529 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1530 // Generic SHUFPD patterns
1531 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1532 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1533 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1534 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1535 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1536 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1537 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1540 let Predicates = [HasAVX] in {
1541 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1542 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1543 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1544 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1545 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1546 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1547 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1548 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1549 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1550 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1551 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1552 // fall back to this for SSE1)
1553 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1554 (VSHUFPSrri VR128:$src2, VR128:$src1,
1555 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1556 // Special unary SHUFPSrri case.
1557 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1558 (VSHUFPSrri VR128:$src1, VR128:$src1,
1559 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1560 // Special binary v4i32 shuffle cases with SHUFPS.
1561 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1562 (VSHUFPSrri VR128:$src1, VR128:$src2,
1563 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1564 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1565 (bc_v4i32 (memopv2i64 addr:$src2)))),
1566 (VSHUFPSrmi VR128:$src1, addr:$src2,
1567 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1568 // Special unary SHUFPDrri cases.
1569 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1570 (VSHUFPDrri VR128:$src1, VR128:$src1,
1571 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1572 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1573 (VSHUFPDrri VR128:$src1, VR128:$src1,
1574 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1575 // Special binary v2i64 shuffle cases using SHUFPDrri.
1576 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1577 (VSHUFPDrri VR128:$src1, VR128:$src2,
1578 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1579 // Generic VSHUFPD patterns
1580 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1581 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1582 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1583 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1584 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1585 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1586 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1589 //===----------------------------------------------------------------------===//
1590 // SSE 1 & 2 - Unpack Instructions
1591 //===----------------------------------------------------------------------===//
1593 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1594 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1595 PatFrag mem_frag, RegisterClass RC,
1596 X86MemOperand x86memop, string asm,
1598 def rr : PI<opc, MRMSrcReg,
1599 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1601 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1602 def rm : PI<opc, MRMSrcMem,
1603 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1605 (vt (OpNode RC:$src1,
1606 (mem_frag addr:$src2))))], d>;
1609 let AddedComplexity = 10 in {
1610 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1611 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1612 SSEPackedSingle>, VEX_4V;
1613 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1614 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1615 SSEPackedDouble>, OpSize, VEX_4V;
1616 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1617 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1618 SSEPackedSingle>, VEX_4V;
1619 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1620 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1621 SSEPackedDouble>, OpSize, VEX_4V;
1623 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1624 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1625 SSEPackedSingle>, VEX_4V;
1626 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1627 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1628 SSEPackedDouble>, OpSize, VEX_4V;
1629 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1630 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1631 SSEPackedSingle>, VEX_4V;
1632 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1633 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1634 SSEPackedDouble>, OpSize, VEX_4V;
1636 let Constraints = "$src1 = $dst" in {
1637 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1638 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1639 SSEPackedSingle>, TB;
1640 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1641 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1642 SSEPackedDouble>, TB, OpSize;
1643 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1644 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1645 SSEPackedSingle>, TB;
1646 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1647 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1648 SSEPackedDouble>, TB, OpSize;
1649 } // Constraints = "$src1 = $dst"
1650 } // AddedComplexity
1652 let Predicates = [HasSSE1] in {
1653 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
1654 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
1655 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
1656 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
1657 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
1658 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
1659 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
1660 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
1663 let Predicates = [HasSSE2] in {
1664 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
1665 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
1666 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
1667 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
1668 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
1669 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
1670 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
1671 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
1673 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
1674 // problem is during lowering, where it's not possible to recognize the load
1675 // fold cause it has two uses through a bitcast. One use disappears at isel
1676 // time and the fold opportunity reappears.
1677 def : Pat<(v2f64 (X86Movddup VR128:$src)),
1678 (UNPCKLPDrr VR128:$src, VR128:$src)>;
1680 let AddedComplexity = 10 in
1681 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
1682 (UNPCKLPDrr VR128:$src, VR128:$src)>;
1685 let Predicates = [HasAVX] in {
1686 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
1687 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
1688 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
1689 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
1690 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
1691 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
1692 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
1693 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
1695 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
1696 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
1697 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
1698 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
1699 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
1700 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
1701 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
1702 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
1703 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
1704 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
1705 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
1706 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
1707 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
1708 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
1709 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
1710 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
1712 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
1713 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
1714 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
1715 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
1716 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
1717 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
1718 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
1719 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
1721 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
1722 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
1723 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
1724 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
1725 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
1726 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
1727 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
1728 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
1729 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
1730 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
1731 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
1732 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
1733 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
1734 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
1735 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
1736 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
1738 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
1739 // problem is during lowering, where it's not possible to recognize the load
1740 // fold cause it has two uses through a bitcast. One use disappears at isel
1741 // time and the fold opportunity reappears.
1742 def : Pat<(v2f64 (X86Movddup VR128:$src)),
1743 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
1744 let AddedComplexity = 10 in
1745 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
1746 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
1749 //===----------------------------------------------------------------------===//
1750 // SSE 1 & 2 - Extract Floating-Point Sign mask
1751 //===----------------------------------------------------------------------===//
1753 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1754 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1756 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1757 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1758 [(set GR32:$dst, (Int RC:$src))], d>;
1759 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1760 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1763 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1764 SSEPackedSingle>, TB;
1765 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1766 SSEPackedDouble>, TB, OpSize;
1768 def : Pat<(i32 (X86fgetsign FR32:$src)),
1769 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1770 sub_ss))>, Requires<[HasSSE1]>;
1771 def : Pat<(i64 (X86fgetsign FR32:$src)),
1772 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1773 sub_ss))>, Requires<[HasSSE1]>;
1774 def : Pat<(i32 (X86fgetsign FR64:$src)),
1775 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1776 sub_sd))>, Requires<[HasSSE2]>;
1777 def : Pat<(i64 (X86fgetsign FR64:$src)),
1778 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1779 sub_sd))>, Requires<[HasSSE2]>;
1781 let Predicates = [HasAVX] in {
1782 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1783 "movmskps", SSEPackedSingle>, TB, VEX;
1784 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1785 "movmskpd", SSEPackedDouble>, TB, OpSize,
1787 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1788 "movmskps", SSEPackedSingle>, TB, VEX;
1789 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1790 "movmskpd", SSEPackedDouble>, TB, OpSize,
1793 def : Pat<(i32 (X86fgetsign FR32:$src)),
1794 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1796 def : Pat<(i64 (X86fgetsign FR32:$src)),
1797 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1799 def : Pat<(i32 (X86fgetsign FR64:$src)),
1800 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1802 def : Pat<(i64 (X86fgetsign FR64:$src)),
1803 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1807 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1808 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1809 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1810 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1812 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1813 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1814 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1815 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1819 //===----------------------------------------------------------------------===//
1820 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1821 //===----------------------------------------------------------------------===//
1823 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1824 // names that start with 'Fs'.
1826 // Alias instructions that map fld0 to pxor for sse.
1827 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1828 canFoldAsLoad = 1 in {
1829 // FIXME: Set encoding to pseudo!
1830 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1831 [(set FR32:$dst, fp32imm0)]>,
1832 Requires<[HasSSE1]>, TB, OpSize;
1833 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1834 [(set FR64:$dst, fpimm0)]>,
1835 Requires<[HasSSE2]>, TB, OpSize;
1836 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1837 [(set FR32:$dst, fp32imm0)]>,
1838 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1839 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1840 [(set FR64:$dst, fpimm0)]>,
1841 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1844 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1845 // bits are disregarded.
1846 let neverHasSideEffects = 1 in {
1847 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1848 "movaps\t{$src, $dst|$dst, $src}", []>;
1849 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1850 "movapd\t{$src, $dst|$dst, $src}", []>;
1853 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1854 // bits are disregarded.
1855 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1856 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1857 "movaps\t{$src, $dst|$dst, $src}",
1858 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1859 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1860 "movapd\t{$src, $dst|$dst, $src}",
1861 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1864 //===----------------------------------------------------------------------===//
1865 // SSE 1 & 2 - Logical Instructions
1866 //===----------------------------------------------------------------------===//
1868 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1870 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1872 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1873 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
1875 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1876 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
1878 let Constraints = "$src1 = $dst" in {
1879 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1880 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1882 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1883 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1887 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1888 let mayLoad = 0 in {
1889 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1890 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1891 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1894 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1895 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1897 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1899 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1901 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
1902 // are all promoted to v2i64, and the patterns are covered by the int
1903 // version. This is needed in SSE only, because v2i64 isn't supported on
1904 // SSE1, but only on SSE2.
1905 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1906 !strconcat(OpcodeStr, "ps"), f128mem, [],
1907 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1908 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
1910 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1911 !strconcat(OpcodeStr, "pd"), f128mem,
1912 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1913 (bc_v2i64 (v2f64 VR128:$src2))))],
1914 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1915 (memopv2i64 addr:$src2)))], 0>,
1917 let Constraints = "$src1 = $dst" in {
1918 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1919 !strconcat(OpcodeStr, "ps"), f128mem,
1920 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
1921 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1922 (memopv2i64 addr:$src2)))]>, TB;
1924 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1925 !strconcat(OpcodeStr, "pd"), f128mem,
1926 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1927 (bc_v2i64 (v2f64 VR128:$src2))))],
1928 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1929 (memopv2i64 addr:$src2)))]>, TB, OpSize;
1933 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1935 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
1937 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1938 !strconcat(OpcodeStr, "ps"), f256mem,
1939 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
1940 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
1941 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
1943 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1944 !strconcat(OpcodeStr, "pd"), f256mem,
1945 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1946 (bc_v4i64 (v4f64 VR256:$src2))))],
1947 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
1948 (memopv4i64 addr:$src2)))], 0>,
1952 // AVX 256-bit packed logical ops forms
1953 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
1954 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
1955 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
1956 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
1958 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1959 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1960 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1961 let isCommutable = 0 in
1962 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
1964 //===----------------------------------------------------------------------===//
1965 // SSE 1 & 2 - Arithmetic Instructions
1966 //===----------------------------------------------------------------------===//
1968 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1971 /// In addition, we also have a special variant of the scalar form here to
1972 /// represent the associated intrinsic operation. This form is unlike the
1973 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1974 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1976 /// These three forms can each be reg+reg or reg+mem.
1979 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1981 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1983 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1984 OpNode, FR32, f32mem, Is2Addr>, XS;
1985 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1986 OpNode, FR64, f64mem, Is2Addr>, XD;
1989 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1991 let mayLoad = 0 in {
1992 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1993 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1994 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1995 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1999 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2001 let mayLoad = 0 in {
2002 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2003 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2004 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2005 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2009 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2011 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2012 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2013 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2014 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2017 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2019 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2020 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2021 SSEPackedSingle, Is2Addr>, TB;
2023 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2024 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2025 SSEPackedDouble, Is2Addr>, TB, OpSize;
2028 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2029 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2030 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2031 SSEPackedSingle, 0>, TB;
2033 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2034 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2035 SSEPackedDouble, 0>, TB, OpSize;
2038 // Binary Arithmetic instructions
2039 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2040 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
2041 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2042 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2043 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2044 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
2045 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2046 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2048 let isCommutable = 0 in {
2049 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2050 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
2051 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2052 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2053 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2054 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
2055 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2056 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2057 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2058 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
2059 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2060 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2061 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2062 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2063 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2064 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
2065 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2066 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2067 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2068 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2071 let Constraints = "$src1 = $dst" in {
2072 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2073 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2074 basic_sse12_fp_binop_s_int<0x58, "add">;
2075 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2076 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2077 basic_sse12_fp_binop_s_int<0x59, "mul">;
2079 let isCommutable = 0 in {
2080 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2081 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2082 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2083 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2084 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2085 basic_sse12_fp_binop_s_int<0x5E, "div">;
2086 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2087 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2088 basic_sse12_fp_binop_s_int<0x5F, "max">,
2089 basic_sse12_fp_binop_p_int<0x5F, "max">;
2090 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2091 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2092 basic_sse12_fp_binop_s_int<0x5D, "min">,
2093 basic_sse12_fp_binop_p_int<0x5D, "min">;
2098 /// In addition, we also have a special variant of the scalar form here to
2099 /// represent the associated intrinsic operation. This form is unlike the
2100 /// plain scalar form, in that it takes an entire vector (instead of a
2101 /// scalar) and leaves the top elements undefined.
2103 /// And, we have a special variant form for a full-vector intrinsic form.
2105 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2106 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2107 SDNode OpNode, Intrinsic F32Int> {
2108 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2109 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2110 [(set FR32:$dst, (OpNode FR32:$src))]>;
2111 // For scalar unary operations, fold a load into the operation
2112 // only in OptForSize mode. It eliminates an instruction, but it also
2113 // eliminates a whole-register clobber (the load), so it introduces a
2114 // partial register update condition.
2115 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2116 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2117 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2118 Requires<[HasSSE1, OptForSize]>;
2119 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2120 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2121 [(set VR128:$dst, (F32Int VR128:$src))]>;
2122 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2123 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2124 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2127 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2128 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2129 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2130 !strconcat(OpcodeStr,
2131 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2132 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2133 !strconcat(OpcodeStr,
2134 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2135 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2136 (ins ssmem:$src1, VR128:$src2),
2137 !strconcat(OpcodeStr,
2138 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2141 /// sse1_fp_unop_p - SSE1 unops in packed form.
2142 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2143 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2144 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2145 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2146 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2147 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2148 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2151 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2152 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2153 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2154 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2155 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2156 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2157 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2158 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2161 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2162 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2163 Intrinsic V4F32Int> {
2164 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2165 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2166 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2167 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2168 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2169 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2172 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2173 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2174 Intrinsic V4F32Int> {
2175 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2176 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2177 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2178 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2179 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2180 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2183 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2184 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2185 SDNode OpNode, Intrinsic F64Int> {
2186 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2187 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2188 [(set FR64:$dst, (OpNode FR64:$src))]>;
2189 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2190 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2191 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2192 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2193 Requires<[HasSSE2, OptForSize]>;
2194 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2195 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2196 [(set VR128:$dst, (F64Int VR128:$src))]>;
2197 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2198 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2199 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2202 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2203 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2204 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2205 !strconcat(OpcodeStr,
2206 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2207 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2208 !strconcat(OpcodeStr,
2209 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2210 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2211 (ins VR128:$src1, sdmem:$src2),
2212 !strconcat(OpcodeStr,
2213 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2216 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2217 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2219 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2220 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2221 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2222 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2223 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2224 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2227 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2228 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2229 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2230 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2231 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2232 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2233 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2234 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2237 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2238 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2239 Intrinsic V2F64Int> {
2240 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2241 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2242 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2243 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2244 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2245 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2248 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2249 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2250 Intrinsic V2F64Int> {
2251 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2252 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2253 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2254 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2255 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2256 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2259 let Predicates = [HasAVX] in {
2261 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2262 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
2264 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2265 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2266 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2267 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2268 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2269 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2270 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2271 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2274 // Reciprocal approximations. Note that these typically require refinement
2275 // in order to obtain suitable precision.
2276 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
2277 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2278 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2279 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2280 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2282 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
2283 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2284 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2285 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2286 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2289 def : Pat<(f32 (fsqrt FR32:$src)),
2290 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2291 def : Pat<(f32 (fsqrt (load addr:$src))),
2292 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2293 Requires<[HasAVX, OptForSize]>;
2294 def : Pat<(f64 (fsqrt FR64:$src)),
2295 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2296 def : Pat<(f64 (fsqrt (load addr:$src))),
2297 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2298 Requires<[HasAVX, OptForSize]>;
2300 def : Pat<(f32 (X86frsqrt FR32:$src)),
2301 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2302 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2303 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2304 Requires<[HasAVX, OptForSize]>;
2306 def : Pat<(f32 (X86frcp FR32:$src)),
2307 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2308 def : Pat<(f32 (X86frcp (load addr:$src))),
2309 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2310 Requires<[HasAVX, OptForSize]>;
2312 let Predicates = [HasAVX] in {
2313 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2314 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2315 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2316 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2318 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2319 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2321 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2322 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2323 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2324 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2326 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2327 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2329 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2330 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2331 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2332 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2334 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2335 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2337 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2338 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2339 (VRCPSSr (f32 (IMPLICIT_DEF)),
2340 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2342 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2343 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2347 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2348 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2349 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2350 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2351 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2352 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2354 // Reciprocal approximations. Note that these typically require refinement
2355 // in order to obtain suitable precision.
2356 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2357 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2358 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2359 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2360 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2361 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2363 // There is no f64 version of the reciprocal approximation instructions.
2365 //===----------------------------------------------------------------------===//
2366 // SSE 1 & 2 - Non-temporal stores
2367 //===----------------------------------------------------------------------===//
2369 let AddedComplexity = 400 in { // Prefer non-temporal versions
2370 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2371 (ins f128mem:$dst, VR128:$src),
2372 "movntps\t{$src, $dst|$dst, $src}",
2373 [(alignednontemporalstore (v4f32 VR128:$src),
2375 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2376 (ins f128mem:$dst, VR128:$src),
2377 "movntpd\t{$src, $dst|$dst, $src}",
2378 [(alignednontemporalstore (v2f64 VR128:$src),
2380 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2381 (ins f128mem:$dst, VR128:$src),
2382 "movntdq\t{$src, $dst|$dst, $src}",
2383 [(alignednontemporalstore (v2f64 VR128:$src),
2386 let ExeDomain = SSEPackedInt in
2387 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2388 (ins f128mem:$dst, VR128:$src),
2389 "movntdq\t{$src, $dst|$dst, $src}",
2390 [(alignednontemporalstore (v4f32 VR128:$src),
2393 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2394 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
2396 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2397 (ins f256mem:$dst, VR256:$src),
2398 "movntps\t{$src, $dst|$dst, $src}",
2399 [(alignednontemporalstore (v8f32 VR256:$src),
2401 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2402 (ins f256mem:$dst, VR256:$src),
2403 "movntpd\t{$src, $dst|$dst, $src}",
2404 [(alignednontemporalstore (v4f64 VR256:$src),
2406 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2407 (ins f256mem:$dst, VR256:$src),
2408 "movntdq\t{$src, $dst|$dst, $src}",
2409 [(alignednontemporalstore (v4f64 VR256:$src),
2411 let ExeDomain = SSEPackedInt in
2412 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2413 (ins f256mem:$dst, VR256:$src),
2414 "movntdq\t{$src, $dst|$dst, $src}",
2415 [(alignednontemporalstore (v8f32 VR256:$src),
2419 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2420 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2421 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2422 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2423 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2424 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2426 let AddedComplexity = 400 in { // Prefer non-temporal versions
2427 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2428 "movntps\t{$src, $dst|$dst, $src}",
2429 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2430 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2431 "movntpd\t{$src, $dst|$dst, $src}",
2432 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2434 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2435 "movntdq\t{$src, $dst|$dst, $src}",
2436 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2438 let ExeDomain = SSEPackedInt in
2439 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2440 "movntdq\t{$src, $dst|$dst, $src}",
2441 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2443 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2444 (MOVNTDQmr addr:$dst, VR128:$src)>;
2446 // There is no AVX form for instructions below this point
2447 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2448 "movnti{l}\t{$src, $dst|$dst, $src}",
2449 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2450 TB, Requires<[HasSSE2]>;
2451 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2452 "movnti{q}\t{$src, $dst|$dst, $src}",
2453 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2454 TB, Requires<[HasSSE2]>;
2457 //===----------------------------------------------------------------------===//
2458 // SSE 1 & 2 - Misc Instructions (No AVX form)
2459 //===----------------------------------------------------------------------===//
2461 // Prefetch intrinsic.
2462 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2463 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2464 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2465 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2466 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2467 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2468 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2469 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2471 // Load, store, and memory fence
2472 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2473 TB, Requires<[HasSSE1]>;
2474 def : Pat<(X86SFence), (SFENCE)>;
2476 // Alias instructions that map zero vector to pxor / xorp* for sse.
2477 // We set canFoldAsLoad because this can be converted to a constant-pool
2478 // load of an all-zeros value if folding it would be beneficial.
2479 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2480 // JIT implementation, it does not expand the instructions below like
2481 // X86MCInstLower does.
2482 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2483 isCodeGenOnly = 1 in {
2484 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2485 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2486 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2487 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2488 let ExeDomain = SSEPackedInt in
2489 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2490 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2493 // The same as done above but for AVX. The 128-bit versions are the
2494 // same, but re-encoded. The 256-bit does not support PI version, and
2495 // doesn't need it because on sandy bridge the register is set to zero
2496 // at the rename stage without using any execution unit, so SET0PSY
2497 // and SET0PDY can be used for vector int instructions without penalty
2498 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2499 // JIT implementatioan, it does not expand the instructions below like
2500 // X86MCInstLower does.
2501 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2502 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2503 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2504 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2505 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2506 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2507 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2508 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2509 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2510 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2511 let ExeDomain = SSEPackedInt in
2512 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2513 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2516 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2517 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2518 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2520 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2521 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2523 // AVX has no support for 256-bit integer instructions, but since the 128-bit
2524 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
2525 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
2526 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
2527 (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
2529 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
2530 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
2531 (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
2533 //===----------------------------------------------------------------------===//
2534 // SSE 1 & 2 - Load/Store XCSR register
2535 //===----------------------------------------------------------------------===//
2537 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2538 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2539 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2540 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2542 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2543 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2544 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2545 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2547 //===---------------------------------------------------------------------===//
2548 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2549 //===---------------------------------------------------------------------===//
2551 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2553 let neverHasSideEffects = 1 in {
2554 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2555 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2556 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2557 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2559 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2560 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2561 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2562 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2564 let canFoldAsLoad = 1, mayLoad = 1 in {
2565 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2566 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2567 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2568 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2569 let Predicates = [HasAVX] in {
2570 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2571 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2572 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2573 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2577 let mayStore = 1 in {
2578 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2579 (ins i128mem:$dst, VR128:$src),
2580 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2581 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2582 (ins i256mem:$dst, VR256:$src),
2583 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2584 let Predicates = [HasAVX] in {
2585 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2586 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2587 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2588 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2592 let neverHasSideEffects = 1 in
2593 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2594 "movdqa\t{$src, $dst|$dst, $src}", []>;
2596 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2597 "movdqu\t{$src, $dst|$dst, $src}",
2598 []>, XS, Requires<[HasSSE2]>;
2600 let canFoldAsLoad = 1, mayLoad = 1 in {
2601 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2602 "movdqa\t{$src, $dst|$dst, $src}",
2603 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2604 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2605 "movdqu\t{$src, $dst|$dst, $src}",
2606 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2607 XS, Requires<[HasSSE2]>;
2610 let mayStore = 1 in {
2611 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2612 "movdqa\t{$src, $dst|$dst, $src}",
2613 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2614 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2615 "movdqu\t{$src, $dst|$dst, $src}",
2616 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2617 XS, Requires<[HasSSE2]>;
2620 // Intrinsic forms of MOVDQU load and store
2621 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2622 "vmovdqu\t{$src, $dst|$dst, $src}",
2623 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2624 XS, VEX, Requires<[HasAVX]>;
2626 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2627 "movdqu\t{$src, $dst|$dst, $src}",
2628 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2629 XS, Requires<[HasSSE2]>;
2631 } // ExeDomain = SSEPackedInt
2633 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2634 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2635 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2637 //===---------------------------------------------------------------------===//
2638 // SSE2 - Packed Integer Arithmetic Instructions
2639 //===---------------------------------------------------------------------===//
2641 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2643 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2644 bit IsCommutable = 0, bit Is2Addr = 1> {
2645 let isCommutable = IsCommutable in
2646 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2647 (ins VR128:$src1, VR128:$src2),
2649 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2650 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2651 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2652 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2653 (ins VR128:$src1, i128mem:$src2),
2655 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2656 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2657 [(set VR128:$dst, (IntId VR128:$src1,
2658 (bitconvert (memopv2i64 addr:$src2))))]>;
2661 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2662 string OpcodeStr, Intrinsic IntId,
2663 Intrinsic IntId2, bit Is2Addr = 1> {
2664 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2665 (ins VR128:$src1, VR128:$src2),
2667 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2668 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2669 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2670 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2671 (ins VR128:$src1, i128mem:$src2),
2673 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2674 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2675 [(set VR128:$dst, (IntId VR128:$src1,
2676 (bitconvert (memopv2i64 addr:$src2))))]>;
2677 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2678 (ins VR128:$src1, i32i8imm:$src2),
2680 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2681 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2682 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2685 /// PDI_binop_rm - Simple SSE2 binary operator.
2686 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2687 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2688 let isCommutable = IsCommutable in
2689 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2690 (ins VR128:$src1, VR128:$src2),
2692 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2693 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2694 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2695 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2696 (ins VR128:$src1, i128mem:$src2),
2698 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2699 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2700 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2701 (bitconvert (memopv2i64 addr:$src2)))))]>;
2704 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2706 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2707 /// to collapse (bitconvert VT to VT) into its operand.
2709 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2710 bit IsCommutable = 0, bit Is2Addr = 1> {
2711 let isCommutable = IsCommutable in
2712 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2713 (ins VR128:$src1, VR128:$src2),
2715 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2716 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2717 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2718 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2719 (ins VR128:$src1, i128mem:$src2),
2721 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2722 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2723 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2726 } // ExeDomain = SSEPackedInt
2728 // 128-bit Integer Arithmetic
2730 let Predicates = [HasAVX] in {
2731 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2732 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2733 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2734 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2735 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2736 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2737 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2738 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2739 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2742 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2744 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2746 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2748 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2750 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2752 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2754 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2756 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2758 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2760 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2762 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2764 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2766 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2768 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2770 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2772 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2774 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2776 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2778 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2782 let Constraints = "$src1 = $dst" in {
2783 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2784 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2785 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2786 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2787 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2788 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2789 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2790 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2791 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2794 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2795 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2796 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2797 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2798 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2799 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2800 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2801 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2802 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2803 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2804 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2805 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2806 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2807 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2808 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2809 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2810 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2811 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2812 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2814 } // Constraints = "$src1 = $dst"
2816 //===---------------------------------------------------------------------===//
2817 // SSE2 - Packed Integer Logical Instructions
2818 //===---------------------------------------------------------------------===//
2820 let Predicates = [HasAVX] in {
2821 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2822 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2824 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2825 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2827 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2828 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2831 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2832 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2834 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2835 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2837 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2838 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2841 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2842 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2844 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2845 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2848 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2849 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2850 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2852 let ExeDomain = SSEPackedInt in {
2853 let neverHasSideEffects = 1 in {
2854 // 128-bit logical shifts.
2855 def VPSLLDQri : PDIi8<0x73, MRM7r,
2856 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2857 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2859 def VPSRLDQri : PDIi8<0x73, MRM3r,
2860 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2861 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2863 // PSRADQri doesn't exist in SSE[1-3].
2865 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2866 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2867 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2869 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
2871 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2872 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2873 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2874 [(set VR128:$dst, (X86andnp VR128:$src1,
2875 (memopv2i64 addr:$src2)))]>, VEX_4V;
2879 let Constraints = "$src1 = $dst" in {
2880 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2881 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2882 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2883 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2884 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2885 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2887 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2888 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2889 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2890 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2891 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2892 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2894 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2895 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2896 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2897 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2899 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2900 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2901 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2903 let ExeDomain = SSEPackedInt in {
2904 let neverHasSideEffects = 1 in {
2905 // 128-bit logical shifts.
2906 def PSLLDQri : PDIi8<0x73, MRM7r,
2907 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2908 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2909 def PSRLDQri : PDIi8<0x73, MRM3r,
2910 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2911 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2912 // PSRADQri doesn't exist in SSE[1-3].
2914 def PANDNrr : PDI<0xDF, MRMSrcReg,
2915 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2916 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2918 def PANDNrm : PDI<0xDF, MRMSrcMem,
2919 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2920 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2922 } // Constraints = "$src1 = $dst"
2924 let Predicates = [HasAVX] in {
2925 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2926 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2927 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2928 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2929 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2930 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2931 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2932 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2933 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2934 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2936 // Shift up / down and insert zero's.
2937 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2938 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2939 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2940 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2943 let Predicates = [HasSSE2] in {
2944 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2945 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2946 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2947 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2948 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2949 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2950 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2951 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2952 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2953 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2955 // Shift up / down and insert zero's.
2956 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2957 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2958 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2959 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2962 //===---------------------------------------------------------------------===//
2963 // SSE2 - Packed Integer Comparison Instructions
2964 //===---------------------------------------------------------------------===//
2966 let Predicates = [HasAVX] in {
2967 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2969 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2971 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2973 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2975 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2977 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2980 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2981 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
2982 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2983 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
2984 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2985 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
2986 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2987 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
2988 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2989 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
2990 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2991 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
2993 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2994 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
2995 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2996 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
2997 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2998 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
2999 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3000 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3001 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3002 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3003 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3004 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3007 let Constraints = "$src1 = $dst" in {
3008 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
3009 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
3010 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
3011 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
3012 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
3013 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
3014 } // Constraints = "$src1 = $dst"
3016 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3017 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3018 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3019 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3020 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3021 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3022 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3023 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3024 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3025 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3026 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3027 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3029 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3030 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3031 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3032 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3033 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3034 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3035 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3036 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3037 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3038 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3039 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3040 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3042 //===---------------------------------------------------------------------===//
3043 // SSE2 - Packed Integer Pack Instructions
3044 //===---------------------------------------------------------------------===//
3046 let Predicates = [HasAVX] in {
3047 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3049 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3051 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3055 let Constraints = "$src1 = $dst" in {
3056 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
3057 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
3058 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
3059 } // Constraints = "$src1 = $dst"
3061 //===---------------------------------------------------------------------===//
3062 // SSE2 - Packed Integer Shuffle Instructions
3063 //===---------------------------------------------------------------------===//
3065 let ExeDomain = SSEPackedInt in {
3066 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3068 def ri : Ii8<0x70, MRMSrcReg,
3069 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3070 !strconcat(OpcodeStr,
3071 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3072 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3074 def mi : Ii8<0x70, MRMSrcMem,
3075 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3076 !strconcat(OpcodeStr,
3077 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3078 [(set VR128:$dst, (vt (pshuf_frag:$src2
3079 (bc_frag (memopv2i64 addr:$src1)),
3082 } // ExeDomain = SSEPackedInt
3084 let Predicates = [HasAVX] in {
3085 let AddedComplexity = 5 in
3086 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
3089 // SSE2 with ImmT == Imm8 and XS prefix.
3090 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3093 // SSE2 with ImmT == Imm8 and XD prefix.
3094 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3097 let AddedComplexity = 5 in
3098 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3099 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3100 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3101 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3102 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3104 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3106 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
3107 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3109 (VPSHUFDmi addr:$src1, imm:$imm)>;
3110 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3111 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
3112 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3113 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
3114 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3115 (VPSHUFHWri VR128:$src, imm:$imm)>;
3116 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3118 (VPSHUFHWmi addr:$src, imm:$imm)>;
3119 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3120 (VPSHUFLWri VR128:$src, imm:$imm)>;
3121 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3123 (VPSHUFLWmi addr:$src, imm:$imm)>;
3126 let Predicates = [HasSSE2] in {
3127 let AddedComplexity = 5 in
3128 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
3130 // SSE2 with ImmT == Imm8 and XS prefix.
3131 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
3133 // SSE2 with ImmT == Imm8 and XD prefix.
3134 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
3136 let AddedComplexity = 5 in
3137 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3138 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3139 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3140 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3141 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3143 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3145 (PSHUFDmi addr:$src1, imm:$imm)>;
3146 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3148 (PSHUFDmi addr:$src1, imm:$imm)>;
3149 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3150 (PSHUFDri VR128:$src1, imm:$imm)>;
3151 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3152 (PSHUFDri VR128:$src1, imm:$imm)>;
3153 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3154 (PSHUFHWri VR128:$src, imm:$imm)>;
3155 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3157 (PSHUFHWmi addr:$src, imm:$imm)>;
3158 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3159 (PSHUFLWri VR128:$src, imm:$imm)>;
3160 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3162 (PSHUFLWmi addr:$src, imm:$imm)>;
3165 //===---------------------------------------------------------------------===//
3166 // SSE2 - Packed Integer Unpack Instructions
3167 //===---------------------------------------------------------------------===//
3169 let ExeDomain = SSEPackedInt in {
3170 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3171 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3172 def rr : PDI<opc, MRMSrcReg,
3173 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3175 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3176 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3177 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3178 def rm : PDI<opc, MRMSrcMem,
3179 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3181 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3182 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3183 [(set VR128:$dst, (OpNode VR128:$src1,
3184 (bc_frag (memopv2i64
3188 let Predicates = [HasAVX] in {
3189 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
3190 bc_v16i8, 0>, VEX_4V;
3191 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
3192 bc_v8i16, 0>, VEX_4V;
3193 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
3194 bc_v4i32, 0>, VEX_4V;
3196 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3197 /// knew to collapse (bitconvert VT to VT) into its operand.
3198 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3199 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3200 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3201 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3202 VR128:$src2)))]>, VEX_4V;
3203 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3204 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3205 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3206 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3207 (memopv2i64 addr:$src2))))]>, VEX_4V;
3209 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
3210 bc_v16i8, 0>, VEX_4V;
3211 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
3212 bc_v8i16, 0>, VEX_4V;
3213 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
3214 bc_v4i32, 0>, VEX_4V;
3216 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3217 /// knew to collapse (bitconvert VT to VT) into its operand.
3218 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3219 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3220 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3221 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3222 VR128:$src2)))]>, VEX_4V;
3223 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3224 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3225 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3226 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3227 (memopv2i64 addr:$src2))))]>, VEX_4V;
3230 let Constraints = "$src1 = $dst" in {
3231 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
3232 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
3233 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
3235 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3236 /// knew to collapse (bitconvert VT to VT) into its operand.
3237 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3238 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3239 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3241 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
3242 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3243 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3244 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3246 (v2i64 (X86Punpcklqdq VR128:$src1,
3247 (memopv2i64 addr:$src2))))]>;
3249 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
3250 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
3251 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
3253 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3254 /// knew to collapse (bitconvert VT to VT) into its operand.
3255 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3256 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3257 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3259 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
3260 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3261 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3262 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3264 (v2i64 (X86Punpckhqdq VR128:$src1,
3265 (memopv2i64 addr:$src2))))]>;
3268 } // ExeDomain = SSEPackedInt
3270 //===---------------------------------------------------------------------===//
3271 // SSE2 - Packed Integer Extract and Insert
3272 //===---------------------------------------------------------------------===//
3274 let ExeDomain = SSEPackedInt in {
3275 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3276 def rri : Ii8<0xC4, MRMSrcReg,
3277 (outs VR128:$dst), (ins VR128:$src1,
3278 GR32:$src2, i32i8imm:$src3),
3280 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3281 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3283 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3284 def rmi : Ii8<0xC4, MRMSrcMem,
3285 (outs VR128:$dst), (ins VR128:$src1,
3286 i16mem:$src2, i32i8imm:$src3),
3288 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3289 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3291 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3296 let Predicates = [HasAVX] in
3297 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3298 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3299 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3300 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3301 imm:$src2))]>, OpSize, VEX;
3302 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3303 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3304 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3305 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3309 let Predicates = [HasAVX] in {
3310 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
3311 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3312 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3313 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3314 []>, OpSize, VEX_4V;
3317 let Constraints = "$src1 = $dst" in
3318 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3320 } // ExeDomain = SSEPackedInt
3322 //===---------------------------------------------------------------------===//
3323 // SSE2 - Packed Mask Creation
3324 //===---------------------------------------------------------------------===//
3326 let ExeDomain = SSEPackedInt in {
3328 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3329 "pmovmskb\t{$src, $dst|$dst, $src}",
3330 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
3331 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
3332 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3333 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3334 "pmovmskb\t{$src, $dst|$dst, $src}",
3335 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
3337 } // ExeDomain = SSEPackedInt
3339 //===---------------------------------------------------------------------===//
3340 // SSE2 - Conditional Store
3341 //===---------------------------------------------------------------------===//
3343 let ExeDomain = SSEPackedInt in {
3346 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
3347 (ins VR128:$src, VR128:$mask),
3348 "maskmovdqu\t{$mask, $src|$src, $mask}",
3349 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
3351 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
3352 (ins VR128:$src, VR128:$mask),
3353 "maskmovdqu\t{$mask, $src|$src, $mask}",
3354 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
3357 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3358 "maskmovdqu\t{$mask, $src|$src, $mask}",
3359 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
3361 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3362 "maskmovdqu\t{$mask, $src|$src, $mask}",
3363 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
3365 } // ExeDomain = SSEPackedInt
3367 //===---------------------------------------------------------------------===//
3368 // SSE2 - Move Doubleword
3369 //===---------------------------------------------------------------------===//
3371 //===---------------------------------------------------------------------===//
3372 // Move Int Doubleword to Packed Double Int
3374 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3375 "movd\t{$src, $dst|$dst, $src}",
3377 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
3378 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3379 "movd\t{$src, $dst|$dst, $src}",
3381 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
3383 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3384 "mov{d|q}\t{$src, $dst|$dst, $src}",
3386 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
3387 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3388 "mov{d|q}\t{$src, $dst|$dst, $src}",
3389 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
3391 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3392 "movd\t{$src, $dst|$dst, $src}",
3394 (v4i32 (scalar_to_vector GR32:$src)))]>;
3395 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3396 "movd\t{$src, $dst|$dst, $src}",
3398 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
3399 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3400 "mov{d|q}\t{$src, $dst|$dst, $src}",
3402 (v2i64 (scalar_to_vector GR64:$src)))]>;
3403 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3404 "mov{d|q}\t{$src, $dst|$dst, $src}",
3405 [(set FR64:$dst, (bitconvert GR64:$src))]>;
3407 //===---------------------------------------------------------------------===//
3408 // Move Int Doubleword to Single Scalar
3410 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3411 "movd\t{$src, $dst|$dst, $src}",
3412 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
3414 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3415 "movd\t{$src, $dst|$dst, $src}",
3416 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
3418 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3419 "movd\t{$src, $dst|$dst, $src}",
3420 [(set FR32:$dst, (bitconvert GR32:$src))]>;
3422 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3423 "movd\t{$src, $dst|$dst, $src}",
3424 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
3426 //===---------------------------------------------------------------------===//
3427 // Move Packed Doubleword Int to Packed Double Int
3429 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3430 "movd\t{$src, $dst|$dst, $src}",
3431 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3433 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
3434 (ins i32mem:$dst, VR128:$src),
3435 "movd\t{$src, $dst|$dst, $src}",
3436 [(store (i32 (vector_extract (v4i32 VR128:$src),
3437 (iPTR 0))), addr:$dst)]>, VEX;
3438 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3439 "movd\t{$src, $dst|$dst, $src}",
3440 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3442 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3443 "movd\t{$src, $dst|$dst, $src}",
3444 [(store (i32 (vector_extract (v4i32 VR128:$src),
3445 (iPTR 0))), addr:$dst)]>;
3447 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3448 "mov{d|q}\t{$src, $dst|$dst, $src}",
3449 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
3451 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
3452 "movq\t{$src, $dst|$dst, $src}",
3453 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
3455 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
3456 "mov{d|q}\t{$src, $dst|$dst, $src}",
3457 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3458 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3459 "movq\t{$src, $dst|$dst, $src}",
3460 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3462 //===---------------------------------------------------------------------===//
3463 // Move Scalar Single to Double Int
3465 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3466 "movd\t{$src, $dst|$dst, $src}",
3467 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3468 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3469 "movd\t{$src, $dst|$dst, $src}",
3470 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3471 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3472 "movd\t{$src, $dst|$dst, $src}",
3473 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3474 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3475 "movd\t{$src, $dst|$dst, $src}",
3476 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3478 //===---------------------------------------------------------------------===//
3479 // Patterns and instructions to describe movd/movq to XMM register zero-extends
3481 let AddedComplexity = 15 in {
3482 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3483 "movd\t{$src, $dst|$dst, $src}",
3484 [(set VR128:$dst, (v4i32 (X86vzmovl
3485 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3487 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3488 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3489 [(set VR128:$dst, (v2i64 (X86vzmovl
3490 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3493 let AddedComplexity = 15 in {
3494 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3495 "movd\t{$src, $dst|$dst, $src}",
3496 [(set VR128:$dst, (v4i32 (X86vzmovl
3497 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3498 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3499 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3500 [(set VR128:$dst, (v2i64 (X86vzmovl
3501 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3504 let AddedComplexity = 20 in {
3505 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3506 "movd\t{$src, $dst|$dst, $src}",
3508 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3509 (loadi32 addr:$src))))))]>,
3511 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3512 "movd\t{$src, $dst|$dst, $src}",
3514 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3515 (loadi32 addr:$src))))))]>;
3517 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3518 (MOVZDI2PDIrm addr:$src)>;
3519 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3520 (MOVZDI2PDIrm addr:$src)>;
3521 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3522 (MOVZDI2PDIrm addr:$src)>;
3525 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
3526 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
3527 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
3528 (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
3529 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
3530 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
3531 (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
3532 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
3534 // These are the correct encodings of the instructions so that we know how to
3535 // read correct assembly, even though we continue to emit the wrong ones for
3536 // compatibility with Darwin's buggy assembler.
3537 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3538 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
3539 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3540 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
3541 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3542 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
3543 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3544 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
3545 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3546 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3547 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3548 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3550 //===---------------------------------------------------------------------===//
3551 // SSE2 - Move Quadword
3552 //===---------------------------------------------------------------------===//
3554 //===---------------------------------------------------------------------===//
3555 // Move Quadword Int to Packed Quadword Int
3557 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3558 "vmovq\t{$src, $dst|$dst, $src}",
3560 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3561 VEX, Requires<[HasAVX]>;
3562 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3563 "movq\t{$src, $dst|$dst, $src}",
3565 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3566 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3568 //===---------------------------------------------------------------------===//
3569 // Move Packed Quadword Int to Quadword Int
3571 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3572 "movq\t{$src, $dst|$dst, $src}",
3573 [(store (i64 (vector_extract (v2i64 VR128:$src),
3574 (iPTR 0))), addr:$dst)]>, VEX;
3575 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3576 "movq\t{$src, $dst|$dst, $src}",
3577 [(store (i64 (vector_extract (v2i64 VR128:$src),
3578 (iPTR 0))), addr:$dst)]>;
3580 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3581 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3583 //===---------------------------------------------------------------------===//
3584 // Store / copy lower 64-bits of a XMM register.
3586 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3587 "movq\t{$src, $dst|$dst, $src}",
3588 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3589 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3590 "movq\t{$src, $dst|$dst, $src}",
3591 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3593 let AddedComplexity = 20 in
3594 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3595 "vmovq\t{$src, $dst|$dst, $src}",
3597 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3598 (loadi64 addr:$src))))))]>,
3599 XS, VEX, Requires<[HasAVX]>;
3601 let AddedComplexity = 20 in {
3602 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3603 "movq\t{$src, $dst|$dst, $src}",
3605 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3606 (loadi64 addr:$src))))))]>,
3607 XS, Requires<[HasSSE2]>;
3609 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3610 (MOVZQI2PQIrm addr:$src)>;
3611 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3612 (MOVZQI2PQIrm addr:$src)>;
3613 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3616 //===---------------------------------------------------------------------===//
3617 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3618 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3620 let AddedComplexity = 15 in
3621 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3622 "vmovq\t{$src, $dst|$dst, $src}",
3623 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3624 XS, VEX, Requires<[HasAVX]>;
3625 let AddedComplexity = 15 in
3626 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3627 "movq\t{$src, $dst|$dst, $src}",
3628 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3629 XS, Requires<[HasSSE2]>;
3631 let AddedComplexity = 20 in
3632 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3633 "vmovq\t{$src, $dst|$dst, $src}",
3634 [(set VR128:$dst, (v2i64 (X86vzmovl
3635 (loadv2i64 addr:$src))))]>,
3636 XS, VEX, Requires<[HasAVX]>;
3637 let AddedComplexity = 20 in {
3638 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3639 "movq\t{$src, $dst|$dst, $src}",
3640 [(set VR128:$dst, (v2i64 (X86vzmovl
3641 (loadv2i64 addr:$src))))]>,
3642 XS, Requires<[HasSSE2]>;
3644 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3645 (MOVZPQILo2PQIrm addr:$src)>;
3648 // Instructions to match in the assembler
3649 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3650 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3651 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3652 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3653 // Recognize "movd" with GR64 destination, but encode as a "movq"
3654 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3655 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3657 // Instructions for the disassembler
3658 // xr = XMM register
3661 let Predicates = [HasAVX] in
3662 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3663 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3664 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3665 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3667 //===---------------------------------------------------------------------===//
3668 // SSE2 - Misc Instructions
3669 //===---------------------------------------------------------------------===//
3672 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3673 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3674 TB, Requires<[HasSSE2]>;
3676 // Load, store, and memory fence
3677 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3678 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3679 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3680 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3681 def : Pat<(X86LFence), (LFENCE)>;
3682 def : Pat<(X86MFence), (MFENCE)>;
3685 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3686 // was introduced with SSE2, it's backward compatible.
3687 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3689 // Alias instructions that map zero vector to pxor / xorp* for sse.
3690 // We set canFoldAsLoad because this can be converted to a constant-pool
3691 // load of an all-ones value if folding it would be beneficial.
3692 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
3693 // JIT implementation, it does not expand the instructions below like
3694 // X86MCInstLower does.
3695 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3696 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3697 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3698 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3699 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3700 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
3701 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3702 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
3704 //===---------------------------------------------------------------------===//
3705 // SSE3 - Conversion Instructions
3706 //===---------------------------------------------------------------------===//
3708 // Convert Packed Double FP to Packed DW Integers
3709 let Predicates = [HasAVX] in {
3710 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3711 // register, but the same isn't true when using memory operands instead.
3712 // Provide other assembly rr and rm forms to address this explicitly.
3713 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3714 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3715 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3716 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3719 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3720 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3721 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3722 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3725 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3726 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3727 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3728 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3731 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3732 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3733 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3734 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3736 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
3737 (VCVTPD2DQYrr VR256:$src)>;
3738 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
3739 (VCVTPD2DQYrm addr:$src)>;
3741 // Convert Packed DW Integers to Packed Double FP
3742 let Predicates = [HasAVX] in {
3743 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3744 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3745 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3746 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3747 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3748 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3749 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3750 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3753 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3754 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3755 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3756 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3758 // AVX 256-bit register conversion intrinsics
3759 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3760 (VCVTDQ2PDYrr VR128:$src)>;
3761 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3762 (VCVTDQ2PDYrm addr:$src)>;
3764 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3765 (VCVTPD2DQYrr VR256:$src)>;
3766 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3767 (VCVTPD2DQYrm addr:$src)>;
3769 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
3770 (VCVTDQ2PDYrr VR128:$src)>;
3771 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
3772 (VCVTDQ2PDYrm addr:$src)>;
3774 //===---------------------------------------------------------------------===//
3775 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
3776 //===---------------------------------------------------------------------===//
3777 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
3778 ValueType vt, RegisterClass RC, PatFrag mem_frag,
3779 X86MemOperand x86memop> {
3780 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
3781 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3782 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
3783 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
3784 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3785 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
3788 let Predicates = [HasAVX] in {
3789 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3790 v4f32, VR128, memopv4f32, f128mem>, VEX;
3791 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3792 v4f32, VR128, memopv4f32, f128mem>, VEX;
3793 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3794 v8f32, VR256, memopv8f32, f256mem>, VEX;
3795 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3796 v8f32, VR256, memopv8f32, f256mem>, VEX;
3798 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
3799 memopv4f32, f128mem>;
3800 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
3801 memopv4f32, f128mem>;
3803 let Predicates = [HasSSE3] in {
3804 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3805 (MOVSHDUPrr VR128:$src)>;
3806 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3807 (MOVSHDUPrm addr:$src)>;
3808 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3809 (MOVSLDUPrr VR128:$src)>;
3810 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3811 (MOVSLDUPrm addr:$src)>;
3814 let Predicates = [HasAVX] in {
3815 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3816 (VMOVSHDUPrr VR128:$src)>;
3817 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3818 (VMOVSHDUPrm addr:$src)>;
3819 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3820 (VMOVSLDUPrr VR128:$src)>;
3821 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3822 (VMOVSLDUPrm addr:$src)>;
3823 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
3824 (VMOVSHDUPYrr VR256:$src)>;
3825 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
3826 (VMOVSHDUPYrm addr:$src)>;
3827 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
3828 (VMOVSLDUPYrr VR256:$src)>;
3829 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
3830 (VMOVSLDUPYrm addr:$src)>;
3833 //===---------------------------------------------------------------------===//
3834 // SSE3 - Replicate Double FP - MOVDDUP
3835 //===---------------------------------------------------------------------===//
3837 multiclass sse3_replicate_dfp<string OpcodeStr> {
3838 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3839 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3840 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3841 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3842 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3844 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3848 // FIXME: Merge with above classe when there're patterns for the ymm version
3849 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3850 let Predicates = [HasAVX] in {
3851 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3852 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3854 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3855 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3860 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3861 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3862 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3864 let Predicates = [HasSSE3] in {
3865 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3867 (MOVDDUPrm addr:$src)>;
3868 let AddedComplexity = 5 in {
3869 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
3870 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3871 (MOVDDUPrm addr:$src)>;
3872 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
3873 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3874 (MOVDDUPrm addr:$src)>;
3876 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
3877 (MOVDDUPrm addr:$src)>;
3878 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
3879 (MOVDDUPrm addr:$src)>;
3880 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
3881 (MOVDDUPrm addr:$src)>;
3882 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
3883 (MOVDDUPrm addr:$src)>;
3884 def : Pat<(X86Movddup (bc_v2f64
3885 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
3886 (MOVDDUPrm addr:$src)>;
3889 let Predicates = [HasAVX] in {
3890 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3892 (VMOVDDUPrm addr:$src)>;
3893 let AddedComplexity = 5 in {
3894 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
3895 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3896 (VMOVDDUPrm addr:$src)>;
3897 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
3898 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3899 (VMOVDDUPrm addr:$src)>;
3901 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
3902 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
3903 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
3904 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
3905 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
3906 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
3907 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
3908 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
3909 def : Pat<(X86Movddup (bc_v2f64
3910 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
3911 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
3914 //===---------------------------------------------------------------------===//
3915 // SSE3 - Move Unaligned Integer
3916 //===---------------------------------------------------------------------===//
3918 let Predicates = [HasAVX] in {
3919 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3920 "vlddqu\t{$src, $dst|$dst, $src}",
3921 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3922 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3923 "vlddqu\t{$src, $dst|$dst, $src}",
3924 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3926 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3927 "lddqu\t{$src, $dst|$dst, $src}",
3928 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3930 //===---------------------------------------------------------------------===//
3931 // SSE3 - Arithmetic
3932 //===---------------------------------------------------------------------===//
3934 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3935 X86MemOperand x86memop, bit Is2Addr = 1> {
3936 def rr : I<0xD0, MRMSrcReg,
3937 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3939 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3940 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3941 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3942 def rm : I<0xD0, MRMSrcMem,
3943 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3945 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3946 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3947 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3950 let Predicates = [HasAVX],
3951 ExeDomain = SSEPackedDouble in {
3952 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3953 f128mem, 0>, TB, XD, VEX_4V;
3954 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3955 f128mem, 0>, TB, OpSize, VEX_4V;
3956 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3957 f256mem, 0>, TB, XD, VEX_4V;
3958 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3959 f256mem, 0>, TB, OpSize, VEX_4V;
3961 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3962 ExeDomain = SSEPackedDouble in {
3963 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3965 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3966 f128mem>, TB, OpSize;
3969 //===---------------------------------------------------------------------===//
3970 // SSE3 Instructions
3971 //===---------------------------------------------------------------------===//
3974 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3975 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3976 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3978 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3979 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3980 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3982 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3984 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3985 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3986 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3988 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3989 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3990 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3992 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3993 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3994 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3996 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3998 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3999 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4000 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4003 let Predicates = [HasAVX] in {
4004 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4005 int_x86_sse3_hadd_ps, 0>, VEX_4V;
4006 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4007 int_x86_sse3_hadd_pd, 0>, VEX_4V;
4008 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4009 int_x86_sse3_hsub_ps, 0>, VEX_4V;
4010 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4011 int_x86_sse3_hsub_pd, 0>, VEX_4V;
4012 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4013 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
4014 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4015 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
4016 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4017 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
4018 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4019 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
4022 let Constraints = "$src1 = $dst" in {
4023 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
4024 int_x86_sse3_hadd_ps>;
4025 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
4026 int_x86_sse3_hadd_pd>;
4027 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
4028 int_x86_sse3_hsub_ps>;
4029 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
4030 int_x86_sse3_hsub_pd>;
4033 //===---------------------------------------------------------------------===//
4034 // SSSE3 - Packed Absolute Instructions
4035 //===---------------------------------------------------------------------===//
4038 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4039 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4040 PatFrag mem_frag128, Intrinsic IntId128> {
4041 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4043 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4044 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4047 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4049 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4052 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
4055 let Predicates = [HasAVX] in {
4056 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
4057 int_x86_ssse3_pabs_b_128>, VEX;
4058 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
4059 int_x86_ssse3_pabs_w_128>, VEX;
4060 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
4061 int_x86_ssse3_pabs_d_128>, VEX;
4064 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
4065 int_x86_ssse3_pabs_b_128>;
4066 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
4067 int_x86_ssse3_pabs_w_128>;
4068 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
4069 int_x86_ssse3_pabs_d_128>;
4071 //===---------------------------------------------------------------------===//
4072 // SSSE3 - Packed Binary Operator Instructions
4073 //===---------------------------------------------------------------------===//
4075 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4076 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4077 PatFrag mem_frag128, Intrinsic IntId128,
4079 let isCommutable = 1 in
4080 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4081 (ins VR128:$src1, VR128:$src2),
4083 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4084 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4085 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4087 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4088 (ins VR128:$src1, i128mem:$src2),
4090 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4091 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4093 (IntId128 VR128:$src1,
4094 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4097 let Predicates = [HasAVX] in {
4098 let isCommutable = 0 in {
4099 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
4100 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
4101 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
4102 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
4103 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
4104 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4105 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
4106 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
4107 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
4108 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
4109 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
4110 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4111 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
4112 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4113 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
4114 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
4115 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
4116 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
4117 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
4118 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
4119 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
4120 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
4122 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
4123 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4126 // None of these have i8 immediate fields.
4127 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4128 let isCommutable = 0 in {
4129 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
4130 int_x86_ssse3_phadd_w_128>;
4131 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
4132 int_x86_ssse3_phadd_d_128>;
4133 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
4134 int_x86_ssse3_phadd_sw_128>;
4135 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
4136 int_x86_ssse3_phsub_w_128>;
4137 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
4138 int_x86_ssse3_phsub_d_128>;
4139 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
4140 int_x86_ssse3_phsub_sw_128>;
4141 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
4142 int_x86_ssse3_pmadd_ub_sw_128>;
4143 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
4144 int_x86_ssse3_pshuf_b_128>;
4145 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
4146 int_x86_ssse3_psign_b_128>;
4147 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
4148 int_x86_ssse3_psign_w_128>;
4149 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
4150 int_x86_ssse3_psign_d_128>;
4152 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
4153 int_x86_ssse3_pmul_hr_sw_128>;
4156 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4157 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
4158 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4159 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
4161 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4162 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4163 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4164 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4165 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4166 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4168 //===---------------------------------------------------------------------===//
4169 // SSSE3 - Packed Align Instruction Patterns
4170 //===---------------------------------------------------------------------===//
4172 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4173 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4174 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4176 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4178 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4180 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4181 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4183 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4185 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4189 let Predicates = [HasAVX] in
4190 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4191 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4192 defm PALIGN : ssse3_palign<"palignr">;
4194 let Predicates = [HasSSSE3] in {
4195 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4196 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4197 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4198 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4199 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4200 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4201 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4202 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4205 let Predicates = [HasAVX] in {
4206 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4207 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4208 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4209 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4210 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4211 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4212 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4213 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4216 //===---------------------------------------------------------------------===//
4217 // SSSE3 Misc Instructions
4218 //===---------------------------------------------------------------------===//
4220 // Thread synchronization
4221 let usesCustomInserter = 1 in {
4222 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
4223 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
4224 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
4225 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
4228 let Uses = [EAX, ECX, EDX] in
4229 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
4230 Requires<[HasSSE3]>;
4231 let Uses = [ECX, EAX] in
4232 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
4233 Requires<[HasSSE3]>;
4235 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
4236 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
4238 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
4239 Requires<[In32BitMode]>;
4240 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
4241 Requires<[In64BitMode]>;
4243 //===---------------------------------------------------------------------===//
4244 // Non-Instruction Patterns
4245 //===---------------------------------------------------------------------===//
4247 // extload f32 -> f64. This matches load+fextend because we have a hack in
4248 // the isel (PreprocessForFPConvert) that can introduce loads after dag
4250 // Since these loads aren't folded into the fextend, we have to match it
4252 let Predicates = [HasSSE2] in
4253 def : Pat<(fextend (loadf32 addr:$src)),
4254 (CVTSS2SDrm addr:$src)>;
4256 // Bitcasts between 128-bit vector types. Return the original type since
4257 // no instruction is needed for the conversion
4258 let Predicates = [HasXMMInt] in {
4259 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
4260 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
4261 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
4262 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
4263 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
4264 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
4265 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
4266 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
4267 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
4268 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
4269 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
4270 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
4271 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
4272 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
4273 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
4274 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
4275 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
4276 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
4277 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
4278 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
4279 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
4280 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
4281 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
4282 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
4283 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
4284 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
4285 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
4286 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
4287 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
4288 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
4291 // Bitcasts between 256-bit vector types. Return the original type since
4292 // no instruction is needed for the conversion
4293 let Predicates = [HasAVX] in {
4294 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
4295 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
4296 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
4297 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
4298 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
4299 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
4300 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
4301 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
4302 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
4303 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
4304 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
4305 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
4306 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
4307 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
4308 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
4309 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
4310 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
4311 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
4312 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
4313 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
4314 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
4315 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
4316 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
4317 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
4318 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
4319 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
4320 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
4321 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
4322 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
4323 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
4326 // Move scalar to XMM zero-extended
4327 // movd to XMM register zero-extends
4328 let AddedComplexity = 15 in {
4329 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
4330 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
4331 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
4332 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
4333 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
4334 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
4335 (MOVSSrr (v4f32 (V_SET0PS)),
4336 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
4337 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
4338 (MOVSSrr (v4i32 (V_SET0PI)),
4339 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
4342 // Splat v2f64 / v2i64
4343 let AddedComplexity = 10 in {
4344 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4345 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4348 let AddedComplexity = 20 in {
4349 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
4350 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
4351 (MOVLPSrm VR128:$src1, addr:$src2)>;
4352 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
4353 (MOVLPDrm VR128:$src1, addr:$src2)>;
4354 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
4355 (MOVLPSrm VR128:$src1, addr:$src2)>;
4356 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
4357 (MOVLPDrm VR128:$src1, addr:$src2)>;
4360 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
4361 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4362 (MOVLPSmr addr:$src1, VR128:$src2)>;
4363 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4364 (MOVLPDmr addr:$src1, VR128:$src2)>;
4365 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
4367 (MOVLPSmr addr:$src1, VR128:$src2)>;
4368 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4369 (MOVLPDmr addr:$src1, VR128:$src2)>;
4371 let AddedComplexity = 15 in {
4372 // Setting the lowest element in the vector.
4373 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
4374 (MOVSSrr (v4i32 VR128:$src1),
4375 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
4376 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
4377 (MOVSDrr (v2i64 VR128:$src1),
4378 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
4380 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
4381 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
4382 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4383 Requires<[HasSSE2]>;
4384 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
4385 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4386 Requires<[HasSSE2]>;
4389 // Set lowest element and zero upper elements.
4390 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4391 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
4393 // Use movaps / movups for SSE integer load / store (one byte shorter).
4394 // The instructions selected below are then converted to MOVDQA/MOVDQU
4395 // during the SSE domain pass.
4396 let Predicates = [HasSSE1] in {
4397 def : Pat<(alignedloadv4i32 addr:$src),
4398 (MOVAPSrm addr:$src)>;
4399 def : Pat<(loadv4i32 addr:$src),
4400 (MOVUPSrm addr:$src)>;
4401 def : Pat<(alignedloadv2i64 addr:$src),
4402 (MOVAPSrm addr:$src)>;
4403 def : Pat<(loadv2i64 addr:$src),
4404 (MOVUPSrm addr:$src)>;
4406 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4407 (MOVAPSmr addr:$dst, VR128:$src)>;
4408 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4409 (MOVAPSmr addr:$dst, VR128:$src)>;
4410 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4411 (MOVAPSmr addr:$dst, VR128:$src)>;
4412 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4413 (MOVAPSmr addr:$dst, VR128:$src)>;
4414 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4415 (MOVUPSmr addr:$dst, VR128:$src)>;
4416 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4417 (MOVUPSmr addr:$dst, VR128:$src)>;
4418 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4419 (MOVUPSmr addr:$dst, VR128:$src)>;
4420 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4421 (MOVUPSmr addr:$dst, VR128:$src)>;
4424 // Use vmovaps/vmovups for AVX integer load/store.
4425 let Predicates = [HasAVX] in {
4426 // 128-bit load/store
4427 def : Pat<(alignedloadv4i32 addr:$src),
4428 (VMOVAPSrm addr:$src)>;
4429 def : Pat<(loadv4i32 addr:$src),
4430 (VMOVUPSrm addr:$src)>;
4431 def : Pat<(alignedloadv2i64 addr:$src),
4432 (VMOVAPSrm addr:$src)>;
4433 def : Pat<(loadv2i64 addr:$src),
4434 (VMOVUPSrm addr:$src)>;
4436 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4437 (VMOVAPSmr addr:$dst, VR128:$src)>;
4438 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4439 (VMOVAPSmr addr:$dst, VR128:$src)>;
4440 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4441 (VMOVAPSmr addr:$dst, VR128:$src)>;
4442 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4443 (VMOVAPSmr addr:$dst, VR128:$src)>;
4444 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4445 (VMOVUPSmr addr:$dst, VR128:$src)>;
4446 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4447 (VMOVUPSmr addr:$dst, VR128:$src)>;
4448 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4449 (VMOVUPSmr addr:$dst, VR128:$src)>;
4450 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4451 (VMOVUPSmr addr:$dst, VR128:$src)>;
4453 // 256-bit load/store
4454 def : Pat<(alignedloadv4i64 addr:$src),
4455 (VMOVAPSYrm addr:$src)>;
4456 def : Pat<(loadv4i64 addr:$src),
4457 (VMOVUPSYrm addr:$src)>;
4458 def : Pat<(alignedloadv8i32 addr:$src),
4459 (VMOVAPSYrm addr:$src)>;
4460 def : Pat<(loadv8i32 addr:$src),
4461 (VMOVUPSYrm addr:$src)>;
4462 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
4463 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4464 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
4465 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4466 def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
4467 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4468 def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
4469 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4470 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
4471 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4472 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
4473 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4474 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
4475 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4476 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
4477 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4480 //===----------------------------------------------------------------------===//
4481 // SSE4.1 - Packed Move with Sign/Zero Extend
4482 //===----------------------------------------------------------------------===//
4484 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4485 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4486 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4487 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4489 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4492 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4496 let Predicates = [HasAVX] in {
4497 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4499 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4501 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4503 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4505 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4507 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4511 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4512 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4513 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4514 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4515 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4516 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4518 // Common patterns involving scalar load.
4519 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4520 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4521 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4522 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4524 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4525 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4526 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4527 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4529 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4530 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4531 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4532 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4534 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4535 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4536 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4537 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4539 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4540 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4541 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4542 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4544 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4545 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4546 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4547 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4550 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4551 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4552 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4553 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4555 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4556 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4558 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4562 let Predicates = [HasAVX] in {
4563 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4565 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4567 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4569 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4573 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4574 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4575 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4576 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4578 // Common patterns involving scalar load
4579 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4580 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4581 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4582 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4584 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4585 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4586 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4587 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4590 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4591 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4592 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4593 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4595 // Expecting a i16 load any extended to i32 value.
4596 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4597 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4598 [(set VR128:$dst, (IntId (bitconvert
4599 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4603 let Predicates = [HasAVX] in {
4604 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4606 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4609 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4610 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4612 // Common patterns involving scalar load
4613 def : Pat<(int_x86_sse41_pmovsxbq
4614 (bitconvert (v4i32 (X86vzmovl
4615 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4616 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4618 def : Pat<(int_x86_sse41_pmovzxbq
4619 (bitconvert (v4i32 (X86vzmovl
4620 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4621 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4623 //===----------------------------------------------------------------------===//
4624 // SSE4.1 - Extract Instructions
4625 //===----------------------------------------------------------------------===//
4627 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4628 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4629 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4630 (ins VR128:$src1, i32i8imm:$src2),
4631 !strconcat(OpcodeStr,
4632 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4633 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4635 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4636 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4637 !strconcat(OpcodeStr,
4638 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4641 // There's an AssertZext in the way of writing the store pattern
4642 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4645 let Predicates = [HasAVX] in {
4646 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4647 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4648 (ins VR128:$src1, i32i8imm:$src2),
4649 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4652 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4655 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4656 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4657 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4658 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4659 !strconcat(OpcodeStr,
4660 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4663 // There's an AssertZext in the way of writing the store pattern
4664 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4667 let Predicates = [HasAVX] in
4668 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4670 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4673 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4674 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4675 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4676 (ins VR128:$src1, i32i8imm:$src2),
4677 !strconcat(OpcodeStr,
4678 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4680 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4681 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4682 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4683 !strconcat(OpcodeStr,
4684 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4685 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4686 addr:$dst)]>, OpSize;
4689 let Predicates = [HasAVX] in
4690 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4692 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4694 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4695 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4696 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4697 (ins VR128:$src1, i32i8imm:$src2),
4698 !strconcat(OpcodeStr,
4699 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4701 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4702 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4703 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4704 !strconcat(OpcodeStr,
4705 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4706 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4707 addr:$dst)]>, OpSize, REX_W;
4710 let Predicates = [HasAVX] in
4711 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4713 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4715 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4717 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4718 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4719 (ins VR128:$src1, i32i8imm:$src2),
4720 !strconcat(OpcodeStr,
4721 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4723 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4725 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4726 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4727 !strconcat(OpcodeStr,
4728 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4729 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4730 addr:$dst)]>, OpSize;
4733 let Predicates = [HasAVX] in {
4734 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4735 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4736 (ins VR128:$src1, i32i8imm:$src2),
4737 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4740 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4742 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4743 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4746 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4747 Requires<[HasSSE41]>;
4749 //===----------------------------------------------------------------------===//
4750 // SSE4.1 - Insert Instructions
4751 //===----------------------------------------------------------------------===//
4753 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4754 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4755 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4757 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4759 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4761 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4762 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4763 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4765 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4767 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4769 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4770 imm:$src3))]>, OpSize;
4773 let Predicates = [HasAVX] in
4774 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4775 let Constraints = "$src1 = $dst" in
4776 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4778 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4779 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4780 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4782 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4784 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4786 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4788 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4789 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4791 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4793 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4795 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4796 imm:$src3)))]>, OpSize;
4799 let Predicates = [HasAVX] in
4800 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4801 let Constraints = "$src1 = $dst" in
4802 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4804 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4805 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4806 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4808 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4810 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4812 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4814 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4815 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4817 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4819 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4821 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4822 imm:$src3)))]>, OpSize;
4825 let Predicates = [HasAVX] in
4826 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4827 let Constraints = "$src1 = $dst" in
4828 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4830 // insertps has a few different modes, there's the first two here below which
4831 // are optimized inserts that won't zero arbitrary elements in the destination
4832 // vector. The next one matches the intrinsic and could zero arbitrary elements
4833 // in the target vector.
4834 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4835 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4836 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
4838 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4840 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4842 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4844 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4845 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
4847 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4849 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4851 (X86insrtps VR128:$src1,
4852 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4853 imm:$src3))]>, OpSize;
4856 let Constraints = "$src1 = $dst" in
4857 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4858 let Predicates = [HasAVX] in
4859 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4861 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4862 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4864 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4865 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4866 Requires<[HasSSE41]>;
4868 //===----------------------------------------------------------------------===//
4869 // SSE4.1 - Round Instructions
4870 //===----------------------------------------------------------------------===//
4872 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4873 X86MemOperand x86memop, RegisterClass RC,
4874 PatFrag mem_frag32, PatFrag mem_frag64,
4875 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4876 // Intrinsic operation, reg.
4877 // Vector intrinsic operation, reg
4878 def PSr : SS4AIi8<opcps, MRMSrcReg,
4879 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4880 !strconcat(OpcodeStr,
4881 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4882 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4885 // Vector intrinsic operation, mem
4886 def PSm : Ii8<opcps, MRMSrcMem,
4887 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4888 !strconcat(OpcodeStr,
4889 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4891 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4893 Requires<[HasSSE41]>;
4895 // Vector intrinsic operation, reg
4896 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4897 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4898 !strconcat(OpcodeStr,
4899 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4900 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4903 // Vector intrinsic operation, mem
4904 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4905 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4906 !strconcat(OpcodeStr,
4907 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4909 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4913 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4914 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4915 // Intrinsic operation, reg.
4916 // Vector intrinsic operation, reg
4917 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4918 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4919 !strconcat(OpcodeStr,
4920 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4923 // Vector intrinsic operation, mem
4924 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4925 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4926 !strconcat(OpcodeStr,
4927 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4928 []>, TA, OpSize, Requires<[HasSSE41]>;
4930 // Vector intrinsic operation, reg
4931 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4932 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4933 !strconcat(OpcodeStr,
4934 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4937 // Vector intrinsic operation, mem
4938 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4939 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4940 !strconcat(OpcodeStr,
4941 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4945 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4948 Intrinsic F64Int, bit Is2Addr = 1> {
4949 // Intrinsic operation, reg.
4950 def SSr : SS4AIi8<opcss, MRMSrcReg,
4951 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4953 !strconcat(OpcodeStr,
4954 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4955 !strconcat(OpcodeStr,
4956 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4957 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4960 // Intrinsic operation, mem.
4961 def SSm : SS4AIi8<opcss, MRMSrcMem,
4962 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4964 !strconcat(OpcodeStr,
4965 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4966 !strconcat(OpcodeStr,
4967 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4969 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4972 // Intrinsic operation, reg.
4973 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4974 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4976 !strconcat(OpcodeStr,
4977 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4978 !strconcat(OpcodeStr,
4979 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4980 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4983 // Intrinsic operation, mem.
4984 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4985 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4987 !strconcat(OpcodeStr,
4988 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4989 !strconcat(OpcodeStr,
4990 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4992 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4996 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4998 // Intrinsic operation, reg.
4999 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
5000 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5001 !strconcat(OpcodeStr,
5002 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5005 // Intrinsic operation, mem.
5006 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
5007 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5008 !strconcat(OpcodeStr,
5009 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5012 // Intrinsic operation, reg.
5013 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
5014 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5015 !strconcat(OpcodeStr,
5016 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5019 // Intrinsic operation, mem.
5020 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
5021 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5022 !strconcat(OpcodeStr,
5023 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5027 // FP round - roundss, roundps, roundsd, roundpd
5028 let Predicates = [HasAVX] in {
5030 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5031 memopv4f32, memopv2f64,
5032 int_x86_sse41_round_ps,
5033 int_x86_sse41_round_pd>, VEX;
5034 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5035 memopv8f32, memopv4f64,
5036 int_x86_avx_round_ps_256,
5037 int_x86_avx_round_pd_256>, VEX;
5038 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5039 int_x86_sse41_round_ss,
5040 int_x86_sse41_round_sd, 0>, VEX_4V;
5042 // Instructions for the assembler
5043 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
5045 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
5047 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
5050 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5051 memopv4f32, memopv2f64,
5052 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5053 let Constraints = "$src1 = $dst" in
5054 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5055 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5057 //===----------------------------------------------------------------------===//
5058 // SSE4.1 - Packed Bit Test
5059 //===----------------------------------------------------------------------===//
5061 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5062 // the intel intrinsic that corresponds to this.
5063 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5064 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5065 "vptest\t{$src2, $src1|$src1, $src2}",
5066 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5068 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5069 "vptest\t{$src2, $src1|$src1, $src2}",
5070 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5073 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5074 "vptest\t{$src2, $src1|$src1, $src2}",
5075 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5077 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5078 "vptest\t{$src2, $src1|$src1, $src2}",
5079 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5083 let Defs = [EFLAGS] in {
5084 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5085 "ptest \t{$src2, $src1|$src1, $src2}",
5086 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5088 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5089 "ptest \t{$src2, $src1|$src1, $src2}",
5090 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5094 // The bit test instructions below are AVX only
5095 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5096 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5097 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5098 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5099 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5100 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5101 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5102 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5106 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5107 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5108 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5109 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5110 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5113 //===----------------------------------------------------------------------===//
5114 // SSE4.1 - Misc Instructions
5115 //===----------------------------------------------------------------------===//
5117 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5118 "popcnt{w}\t{$src, $dst|$dst, $src}",
5119 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
5120 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5121 "popcnt{w}\t{$src, $dst|$dst, $src}",
5122 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
5124 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5125 "popcnt{l}\t{$src, $dst|$dst, $src}",
5126 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
5127 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5128 "popcnt{l}\t{$src, $dst|$dst, $src}",
5129 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
5131 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5132 "popcnt{q}\t{$src, $dst|$dst, $src}",
5133 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
5134 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5135 "popcnt{q}\t{$src, $dst|$dst, $src}",
5136 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
5140 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5141 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5142 Intrinsic IntId128> {
5143 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5145 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5146 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5147 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5149 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5152 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
5155 let Predicates = [HasAVX] in
5156 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5157 int_x86_sse41_phminposuw>, VEX;
5158 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5159 int_x86_sse41_phminposuw>;
5161 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5162 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5163 Intrinsic IntId128, bit Is2Addr = 1> {
5164 let isCommutable = 1 in
5165 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5166 (ins VR128:$src1, VR128:$src2),
5168 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5169 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5170 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5171 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5172 (ins VR128:$src1, i128mem:$src2),
5174 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5175 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5177 (IntId128 VR128:$src1,
5178 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5181 let Predicates = [HasAVX] in {
5182 let isCommutable = 0 in
5183 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5185 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
5187 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5189 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5191 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5193 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5195 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5197 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5199 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5201 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5203 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5206 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5207 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
5208 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5209 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
5212 let Constraints = "$src1 = $dst" in {
5213 let isCommutable = 0 in
5214 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5215 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
5216 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5217 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5218 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5219 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5220 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5221 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5222 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5223 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5224 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5227 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5228 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
5229 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5230 (PCMPEQQrm VR128:$src1, addr:$src2)>;
5232 /// SS48I_binop_rm - Simple SSE41 binary operator.
5233 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5234 ValueType OpVT, bit Is2Addr = 1> {
5235 let isCommutable = 1 in
5236 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5237 (ins VR128:$src1, VR128:$src2),
5239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5241 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
5243 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5244 (ins VR128:$src1, i128mem:$src2),
5246 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5247 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5248 [(set VR128:$dst, (OpNode VR128:$src1,
5249 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
5253 let Predicates = [HasAVX] in
5254 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
5255 let Constraints = "$src1 = $dst" in
5256 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
5258 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5259 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5260 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5261 X86MemOperand x86memop, bit Is2Addr = 1> {
5262 let isCommutable = 1 in
5263 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5264 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5266 !strconcat(OpcodeStr,
5267 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5268 !strconcat(OpcodeStr,
5269 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5270 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5272 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5273 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5275 !strconcat(OpcodeStr,
5276 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5277 !strconcat(OpcodeStr,
5278 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5281 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
5285 let Predicates = [HasAVX] in {
5286 let isCommutable = 0 in {
5287 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
5288 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5289 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
5290 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5291 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
5292 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5293 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
5294 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5295 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
5296 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5297 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
5298 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5300 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
5301 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5302 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
5303 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5304 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
5305 VR256, memopv32i8, i256mem, 0>, VEX_4V;
5308 let Constraints = "$src1 = $dst" in {
5309 let isCommutable = 0 in {
5310 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
5311 VR128, memopv16i8, i128mem>;
5312 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
5313 VR128, memopv16i8, i128mem>;
5314 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
5315 VR128, memopv16i8, i128mem>;
5316 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
5317 VR128, memopv16i8, i128mem>;
5319 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
5320 VR128, memopv16i8, i128mem>;
5321 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
5322 VR128, memopv16i8, i128mem>;
5325 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
5326 let Predicates = [HasAVX] in {
5327 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
5328 RegisterClass RC, X86MemOperand x86memop,
5329 PatFrag mem_frag, Intrinsic IntId> {
5330 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
5331 (ins RC:$src1, RC:$src2, RC:$src3),
5332 !strconcat(OpcodeStr,
5333 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5334 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
5335 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5337 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
5338 (ins RC:$src1, x86memop:$src2, RC:$src3),
5339 !strconcat(OpcodeStr,
5340 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5342 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5344 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5348 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5349 memopv16i8, int_x86_sse41_blendvpd>;
5350 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5351 memopv16i8, int_x86_sse41_blendvps>;
5352 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5353 memopv16i8, int_x86_sse41_pblendvb>;
5354 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5355 memopv32i8, int_x86_avx_blendv_pd_256>;
5356 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5357 memopv32i8, int_x86_avx_blendv_ps_256>;
5359 /// SS41I_ternary_int - SSE 4.1 ternary operator
5360 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
5361 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5362 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5363 (ins VR128:$src1, VR128:$src2),
5364 !strconcat(OpcodeStr,
5365 "\t{$src2, $dst|$dst, $src2}"),
5366 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
5369 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5370 (ins VR128:$src1, i128mem:$src2),
5371 !strconcat(OpcodeStr,
5372 "\t{$src2, $dst|$dst, $src2}"),
5375 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
5379 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
5380 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
5381 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
5383 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
5384 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
5386 let Predicates = [HasAVX] in
5387 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5388 "vmovntdqa\t{$src, $dst|$dst, $src}",
5389 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5391 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5392 "movntdqa\t{$src, $dst|$dst, $src}",
5393 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5396 //===----------------------------------------------------------------------===//
5397 // SSE4.2 - Compare Instructions
5398 //===----------------------------------------------------------------------===//
5400 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
5401 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
5402 Intrinsic IntId128, bit Is2Addr = 1> {
5403 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
5404 (ins VR128:$src1, VR128:$src2),
5406 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5408 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5410 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
5411 (ins VR128:$src1, i128mem:$src2),
5413 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5414 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5416 (IntId128 VR128:$src1,
5417 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5420 let Predicates = [HasAVX] in {
5421 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
5424 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5425 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
5426 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5427 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
5430 let Constraints = "$src1 = $dst" in
5431 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
5433 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5434 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
5435 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5436 (PCMPGTQrm VR128:$src1, addr:$src2)>;
5438 //===----------------------------------------------------------------------===//
5439 // SSE4.2 - String/text Processing Instructions
5440 //===----------------------------------------------------------------------===//
5442 // Packed Compare Implicit Length Strings, Return Mask
5443 multiclass pseudo_pcmpistrm<string asm> {
5444 def REG : PseudoI<(outs VR128:$dst),
5445 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5446 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
5448 def MEM : PseudoI<(outs VR128:$dst),
5449 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5450 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
5451 VR128:$src1, (load addr:$src2), imm:$src3))]>;
5454 let Defs = [EFLAGS], usesCustomInserter = 1 in {
5455 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
5456 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
5459 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
5460 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5461 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5462 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5463 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5464 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5465 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5468 let Defs = [XMM0, EFLAGS] in {
5469 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5470 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5471 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5472 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5473 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5474 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5477 // Packed Compare Explicit Length Strings, Return Mask
5478 multiclass pseudo_pcmpestrm<string asm> {
5479 def REG : PseudoI<(outs VR128:$dst),
5480 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5481 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5482 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5483 def MEM : PseudoI<(outs VR128:$dst),
5484 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5485 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5486 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5489 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5490 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5491 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5494 let Predicates = [HasAVX],
5495 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5496 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5497 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5498 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5499 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5500 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5501 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5504 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5505 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5506 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5507 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5508 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5509 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5510 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5513 // Packed Compare Implicit Length Strings, Return Index
5514 let Defs = [ECX, EFLAGS] in {
5515 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5516 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5517 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5518 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5519 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5520 (implicit EFLAGS)]>, OpSize;
5521 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5522 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5523 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5524 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5525 (implicit EFLAGS)]>, OpSize;
5529 let Predicates = [HasAVX] in {
5530 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5532 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5534 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5536 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5538 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5540 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5544 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5545 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5546 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5547 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5548 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5549 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5551 // Packed Compare Explicit Length Strings, Return Index
5552 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5553 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5554 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5555 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5556 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5557 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5558 (implicit EFLAGS)]>, OpSize;
5559 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5560 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5561 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5563 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5564 (implicit EFLAGS)]>, OpSize;
5568 let Predicates = [HasAVX] in {
5569 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5571 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5573 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5575 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5577 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5579 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5583 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5584 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5585 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5586 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5587 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5588 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5590 //===----------------------------------------------------------------------===//
5591 // SSE4.2 - CRC Instructions
5592 //===----------------------------------------------------------------------===//
5594 // No CRC instructions have AVX equivalents
5596 // crc intrinsic instruction
5597 // This set of instructions are only rm, the only difference is the size
5599 let Constraints = "$src1 = $dst" in {
5600 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5601 (ins GR32:$src1, i8mem:$src2),
5602 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5604 (int_x86_sse42_crc32_32_8 GR32:$src1,
5605 (load addr:$src2)))]>;
5606 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5607 (ins GR32:$src1, GR8:$src2),
5608 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5610 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5611 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5612 (ins GR32:$src1, i16mem:$src2),
5613 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5615 (int_x86_sse42_crc32_32_16 GR32:$src1,
5616 (load addr:$src2)))]>,
5618 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5619 (ins GR32:$src1, GR16:$src2),
5620 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5622 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5624 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5625 (ins GR32:$src1, i32mem:$src2),
5626 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5628 (int_x86_sse42_crc32_32_32 GR32:$src1,
5629 (load addr:$src2)))]>;
5630 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5631 (ins GR32:$src1, GR32:$src2),
5632 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5634 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5635 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5636 (ins GR64:$src1, i8mem:$src2),
5637 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5639 (int_x86_sse42_crc32_64_8 GR64:$src1,
5640 (load addr:$src2)))]>,
5642 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5643 (ins GR64:$src1, GR8:$src2),
5644 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5646 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5648 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5649 (ins GR64:$src1, i64mem:$src2),
5650 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5652 (int_x86_sse42_crc32_64_64 GR64:$src1,
5653 (load addr:$src2)))]>,
5655 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5656 (ins GR64:$src1, GR64:$src2),
5657 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5659 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5663 //===----------------------------------------------------------------------===//
5664 // AES-NI Instructions
5665 //===----------------------------------------------------------------------===//
5667 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5668 Intrinsic IntId128, bit Is2Addr = 1> {
5669 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5670 (ins VR128:$src1, VR128:$src2),
5672 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5673 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5674 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5676 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5677 (ins VR128:$src1, i128mem:$src2),
5679 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5680 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5682 (IntId128 VR128:$src1,
5683 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5686 // Perform One Round of an AES Encryption/Decryption Flow
5687 let Predicates = [HasAVX, HasAES] in {
5688 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5689 int_x86_aesni_aesenc, 0>, VEX_4V;
5690 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5691 int_x86_aesni_aesenclast, 0>, VEX_4V;
5692 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5693 int_x86_aesni_aesdec, 0>, VEX_4V;
5694 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5695 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5698 let Constraints = "$src1 = $dst" in {
5699 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5700 int_x86_aesni_aesenc>;
5701 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5702 int_x86_aesni_aesenclast>;
5703 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5704 int_x86_aesni_aesdec>;
5705 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5706 int_x86_aesni_aesdeclast>;
5709 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5710 (AESENCrr VR128:$src1, VR128:$src2)>;
5711 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5712 (AESENCrm VR128:$src1, addr:$src2)>;
5713 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5714 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5715 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5716 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5717 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5718 (AESDECrr VR128:$src1, VR128:$src2)>;
5719 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5720 (AESDECrm VR128:$src1, addr:$src2)>;
5721 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5722 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5723 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5724 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5726 // Perform the AES InvMixColumn Transformation
5727 let Predicates = [HasAVX, HasAES] in {
5728 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5730 "vaesimc\t{$src1, $dst|$dst, $src1}",
5732 (int_x86_aesni_aesimc VR128:$src1))]>,
5734 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5735 (ins i128mem:$src1),
5736 "vaesimc\t{$src1, $dst|$dst, $src1}",
5738 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5741 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5743 "aesimc\t{$src1, $dst|$dst, $src1}",
5745 (int_x86_aesni_aesimc VR128:$src1))]>,
5747 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5748 (ins i128mem:$src1),
5749 "aesimc\t{$src1, $dst|$dst, $src1}",
5751 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5754 // AES Round Key Generation Assist
5755 let Predicates = [HasAVX, HasAES] in {
5756 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5757 (ins VR128:$src1, i8imm:$src2),
5758 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5760 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5762 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5763 (ins i128mem:$src1, i8imm:$src2),
5764 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5766 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5770 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5771 (ins VR128:$src1, i8imm:$src2),
5772 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5774 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5776 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5777 (ins i128mem:$src1, i8imm:$src2),
5778 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5780 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5784 //===----------------------------------------------------------------------===//
5785 // CLMUL Instructions
5786 //===----------------------------------------------------------------------===//
5788 // Carry-less Multiplication instructions
5789 let Constraints = "$src1 = $dst" in {
5790 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5791 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5792 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5795 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5796 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5797 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5801 // AVX carry-less Multiplication instructions
5802 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5803 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5804 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5807 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5808 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5809 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5813 multiclass pclmul_alias<string asm, int immop> {
5814 def : InstAlias<!strconcat("pclmul", asm,
5815 "dq {$src, $dst|$dst, $src}"),
5816 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5818 def : InstAlias<!strconcat("pclmul", asm,
5819 "dq {$src, $dst|$dst, $src}"),
5820 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5822 def : InstAlias<!strconcat("vpclmul", asm,
5823 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5824 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5826 def : InstAlias<!strconcat("vpclmul", asm,
5827 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5828 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5830 defm : pclmul_alias<"hqhq", 0x11>;
5831 defm : pclmul_alias<"hqlq", 0x01>;
5832 defm : pclmul_alias<"lqhq", 0x10>;
5833 defm : pclmul_alias<"lqlq", 0x00>;
5835 //===----------------------------------------------------------------------===//
5837 //===----------------------------------------------------------------------===//
5839 //===----------------------------------------------------------------------===//
5840 // VBROADCAST - Load from memory and broadcast to all elements of the
5841 // destination operand
5843 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5844 X86MemOperand x86memop, Intrinsic Int> :
5845 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5846 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5847 [(set RC:$dst, (Int addr:$src))]>, VEX;
5849 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5850 int_x86_avx_vbroadcastss>;
5851 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5852 int_x86_avx_vbroadcastss_256>;
5853 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5854 int_x86_avx_vbroadcast_sd_256>;
5855 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5856 int_x86_avx_vbroadcastf128_pd_256>;
5858 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5859 (VBROADCASTF128 addr:$src)>;
5861 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
5862 (VBROADCASTSSY addr:$src)>;
5863 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
5864 (VBROADCASTSD addr:$src)>;
5865 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
5866 (VBROADCASTSSY addr:$src)>;
5867 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
5868 (VBROADCASTSD addr:$src)>;
5870 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
5871 (VBROADCASTSS addr:$src)>;
5872 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
5873 (VBROADCASTSS addr:$src)>;
5875 //===----------------------------------------------------------------------===//
5876 // VINSERTF128 - Insert packed floating-point values
5878 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5879 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5880 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5882 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5883 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5884 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5887 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5888 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5889 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5890 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5891 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5892 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5894 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5896 (VINSERTF128rr VR256:$src1, VR128:$src2,
5897 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5898 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5900 (VINSERTF128rr VR256:$src1, VR128:$src2,
5901 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5902 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5904 (VINSERTF128rr VR256:$src1, VR128:$src2,
5905 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5906 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5908 (VINSERTF128rr VR256:$src1, VR128:$src2,
5909 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5910 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
5912 (VINSERTF128rr VR256:$src1, VR128:$src2,
5913 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5914 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
5916 (VINSERTF128rr VR256:$src1, VR128:$src2,
5917 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5919 // Special COPY patterns
5920 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
5921 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5922 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
5923 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5924 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
5925 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5926 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
5927 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5928 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
5929 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5930 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
5931 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
5933 //===----------------------------------------------------------------------===//
5934 // VEXTRACTF128 - Extract packed floating-point values
5936 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5937 (ins VR256:$src1, i8imm:$src2),
5938 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5940 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5941 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5942 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5945 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5946 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5947 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5948 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5949 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5950 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5952 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5953 (v4f32 (VEXTRACTF128rr
5954 (v8f32 VR256:$src1),
5955 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5956 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5957 (v2f64 (VEXTRACTF128rr
5958 (v4f64 VR256:$src1),
5959 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5960 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5961 (v4i32 (VEXTRACTF128rr
5962 (v8i32 VR256:$src1),
5963 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5964 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5965 (v2i64 (VEXTRACTF128rr
5966 (v4i64 VR256:$src1),
5967 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5968 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5969 (v8i16 (VEXTRACTF128rr
5970 (v16i16 VR256:$src1),
5971 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5972 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
5973 (v16i8 (VEXTRACTF128rr
5974 (v32i8 VR256:$src1),
5975 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
5977 // Special COPY patterns
5978 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
5979 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
5980 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
5981 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
5983 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
5984 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
5985 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
5986 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
5988 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
5989 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
5990 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
5991 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
5994 //===----------------------------------------------------------------------===//
5995 // VMASKMOV - Conditional SIMD Packed Loads and Stores
5997 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5998 Intrinsic IntLd, Intrinsic IntLd256,
5999 Intrinsic IntSt, Intrinsic IntSt256,
6000 PatFrag pf128, PatFrag pf256> {
6001 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6002 (ins VR128:$src1, f128mem:$src2),
6003 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6004 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6006 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6007 (ins VR256:$src1, f256mem:$src2),
6008 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6009 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6011 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6012 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6013 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6014 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6015 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6016 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6017 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6018 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6021 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6022 int_x86_avx_maskload_ps,
6023 int_x86_avx_maskload_ps_256,
6024 int_x86_avx_maskstore_ps,
6025 int_x86_avx_maskstore_ps_256,
6026 memopv4f32, memopv8f32>;
6027 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6028 int_x86_avx_maskload_pd,
6029 int_x86_avx_maskload_pd_256,
6030 int_x86_avx_maskstore_pd,
6031 int_x86_avx_maskstore_pd_256,
6032 memopv2f64, memopv4f64>;
6034 //===----------------------------------------------------------------------===//
6035 // VPERMIL - Permute Single and Double Floating-Point Values
6037 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6038 RegisterClass RC, X86MemOperand x86memop_f,
6039 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
6040 Intrinsic IntVar, Intrinsic IntImm> {
6041 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6042 (ins RC:$src1, RC:$src2),
6043 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6044 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6045 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6046 (ins RC:$src1, x86memop_i:$src2),
6047 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6048 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
6050 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6051 (ins RC:$src1, i8imm:$src2),
6052 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6053 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
6054 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6055 (ins x86memop_f:$src1, i8imm:$src2),
6056 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6057 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
6060 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6061 memopv4f32, memopv4i32,
6062 int_x86_avx_vpermilvar_ps,
6063 int_x86_avx_vpermil_ps>;
6064 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6065 memopv8f32, memopv8i32,
6066 int_x86_avx_vpermilvar_ps_256,
6067 int_x86_avx_vpermil_ps_256>;
6068 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6069 memopv2f64, memopv2i64,
6070 int_x86_avx_vpermilvar_pd,
6071 int_x86_avx_vpermil_pd>;
6072 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6073 memopv4f64, memopv4i64,
6074 int_x86_avx_vpermilvar_pd_256,
6075 int_x86_avx_vpermil_pd_256>;
6077 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6078 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6079 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6080 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6081 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6082 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6083 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6084 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6086 //===----------------------------------------------------------------------===//
6087 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6089 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6090 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6091 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6093 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6094 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6095 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6098 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
6099 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6100 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
6101 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6102 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
6103 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6105 def : Pat<(int_x86_avx_vperm2f128_ps_256
6106 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
6107 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6108 def : Pat<(int_x86_avx_vperm2f128_pd_256
6109 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
6110 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6111 def : Pat<(int_x86_avx_vperm2f128_si_256
6112 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
6113 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6115 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6116 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6117 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6118 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6119 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6120 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6121 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6122 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6123 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6124 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6125 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6126 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6128 //===----------------------------------------------------------------------===//
6129 // VZERO - Zero YMM registers
6131 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6132 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6133 // Zero All YMM registers
6134 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6135 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
6139 // Zero Upper bits of YMM registers
6140 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6141 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
6143 //===----------------------------------------------------------------------===//
6144 // SSE Shuffle pattern fragments
6145 //===----------------------------------------------------------------------===//
6147 // This is part of a "work in progress" refactoring. The idea is that all
6148 // vector shuffles are going to be translated into target specific nodes and
6149 // directly matched by the patterns below (which can be changed along the way)
6150 // The AVX version of some but not all of them are described here, and more
6151 // should come in a near future.
6153 // Shuffle with MOVLHPD
6154 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
6155 (scalar_to_vector (loadf64 addr:$src2)))),
6156 (MOVHPDrm VR128:$src1, addr:$src2)>;
6158 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
6159 // is during lowering, where it's not possible to recognize the load fold cause
6160 // it has two uses through a bitcast. One use disappears at isel time and the
6161 // fold opportunity reappears.
6162 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
6163 (scalar_to_vector (loadf64 addr:$src2)))),
6164 (MOVHPDrm VR128:$src1, addr:$src2)>;
6166 // Shuffle with MOVSS
6167 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
6168 (MOVSSrr VR128:$src1, FR32:$src2)>;
6169 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
6170 (MOVSSrr (v4i32 VR128:$src1),
6171 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
6172 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
6173 (MOVSSrr (v4f32 VR128:$src1),
6174 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
6176 // Shuffle with MOVSD
6177 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
6178 (MOVSDrr VR128:$src1, FR64:$src2)>;
6179 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
6180 (MOVSDrr (v2i64 VR128:$src1),
6181 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
6182 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
6183 (MOVSDrr (v2f64 VR128:$src1),
6184 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
6185 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
6186 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6187 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
6188 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6190 // Shuffle with MOVLPS
6191 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
6192 (MOVLPSrm VR128:$src1, addr:$src2)>;
6193 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
6194 (MOVLPSrm VR128:$src1, addr:$src2)>;
6195 def : Pat<(X86Movlps VR128:$src1,
6196 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
6197 (MOVLPSrm VR128:$src1, addr:$src2)>;
6198 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
6199 // is during lowering, where it's not possible to recognize the load fold cause
6200 // it has two uses through a bitcast. One use disappears at isel time and the
6201 // fold opportunity reappears.
6202 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
6203 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6205 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
6206 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6208 // Shuffle with MOVLPD
6209 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6210 (MOVLPDrm VR128:$src1, addr:$src2)>;
6211 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6212 (MOVLPDrm VR128:$src1, addr:$src2)>;
6213 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
6214 (scalar_to_vector (loadf64 addr:$src2)))),
6215 (MOVLPDrm VR128:$src1, addr:$src2)>;
6217 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
6218 def : Pat<(store (f64 (vector_extract
6219 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6220 (MOVHPSmr addr:$dst, VR128:$src)>;
6221 def : Pat<(store (f64 (vector_extract
6222 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6223 (MOVHPDmr addr:$dst, VR128:$src)>;
6225 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
6226 (MOVLPSmr addr:$src1, VR128:$src2)>;
6227 def : Pat<(store (v4i32 (X86Movlps
6228 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
6229 (MOVLPSmr addr:$src1, VR128:$src2)>;
6231 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6232 (MOVLPDmr addr:$src1, VR128:$src2)>;
6233 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6234 (MOVLPDmr addr:$src1, VR128:$src2)>;