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:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 let isAsmParserOnly = 1 in {
139 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
140 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
141 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
142 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
144 let canFoldAsLoad = 1, isReMaterializable = 1 in {
145 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
147 let AddedComplexity = 20 in
148 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
152 let Constraints = "$src1 = $dst" in {
153 def MOVSSrr : sse12_move_rr<FR32, v4f32,
154 "movss\t{$src2, $dst|$dst, $src2}">, XS;
155 def MOVSDrr : sse12_move_rr<FR64, v2f64,
156 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
159 let canFoldAsLoad = 1, isReMaterializable = 1 in {
160 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
162 let AddedComplexity = 20 in
163 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
166 let AddedComplexity = 15 in {
167 // Extract the low 32-bit value from one vector and insert it into another.
168 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
169 (MOVSSrr (v4f32 VR128:$src1),
170 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
171 // Extract the low 64-bit value from one vector and insert it into another.
172 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
173 (MOVSDrr (v2f64 VR128:$src1),
174 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
177 // Implicitly promote a 32-bit scalar to a vector.
178 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
179 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
180 // Implicitly promote a 64-bit scalar to a vector.
181 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
182 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
183 // Implicitly promote a 32-bit scalar to a vector.
184 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
185 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
186 // Implicitly promote a 64-bit scalar to a vector.
187 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
188 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
190 let AddedComplexity = 20 in {
191 // MOVSSrm zeros the high parts of the register; represent this
192 // with SUBREG_TO_REG.
193 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
198 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
199 // MOVSDrm zeros the high parts of the register; represent this
200 // with SUBREG_TO_REG.
201 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
209 def : Pat<(v2f64 (X86vzload addr:$src)),
210 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
213 // Store scalar value to memory.
214 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
215 "movss\t{$src, $dst|$dst, $src}",
216 [(store FR32:$src, addr:$dst)]>;
217 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
218 "movsd\t{$src, $dst|$dst, $src}",
219 [(store FR64:$src, addr:$dst)]>;
221 let isAsmParserOnly = 1 in {
222 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
223 "movss\t{$src, $dst|$dst, $src}",
224 [(store FR32:$src, addr:$dst)]>, XS, VEX;
225 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
226 "movsd\t{$src, $dst|$dst, $src}",
227 [(store FR64:$src, addr:$dst)]>, XD, VEX;
230 // Extract and store.
231 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
235 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
238 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
240 // Move Aligned/Unaligned floating point values
241 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
242 X86MemOperand x86memop, PatFrag ld_frag,
243 string asm, Domain d,
244 bit IsReMaterializable = 1> {
245 let neverHasSideEffects = 1 in
246 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
247 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
248 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
249 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
250 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
251 [(set RC:$dst, (ld_frag addr:$src))], d>;
254 let isAsmParserOnly = 1 in {
255 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
256 "movaps", SSEPackedSingle>, VEX;
257 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
258 "movapd", SSEPackedDouble>, OpSize, VEX;
259 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
260 "movups", SSEPackedSingle>, VEX;
261 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
262 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
264 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
265 "movaps", SSEPackedSingle>, VEX;
266 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
267 "movapd", SSEPackedDouble>, OpSize, VEX;
268 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
269 "movups", SSEPackedSingle>, VEX;
270 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
271 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
273 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
274 "movaps", SSEPackedSingle>, TB;
275 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
276 "movapd", SSEPackedDouble>, TB, OpSize;
277 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
278 "movups", SSEPackedSingle>, TB;
279 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
280 "movupd", SSEPackedDouble, 0>, TB, OpSize;
282 let isAsmParserOnly = 1 in {
283 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
284 "movaps\t{$src, $dst|$dst, $src}",
285 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
286 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
287 "movapd\t{$src, $dst|$dst, $src}",
288 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
289 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
290 "movups\t{$src, $dst|$dst, $src}",
291 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
292 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
293 "movupd\t{$src, $dst|$dst, $src}",
294 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
295 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
296 "movaps\t{$src, $dst|$dst, $src}",
297 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
298 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
299 "movapd\t{$src, $dst|$dst, $src}",
300 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
302 "movups\t{$src, $dst|$dst, $src}",
303 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
304 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
305 "movupd\t{$src, $dst|$dst, $src}",
306 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
309 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
310 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
311 (VMOVUPSYmr addr:$dst, VR256:$src)>;
313 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
314 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
315 (VMOVUPDYmr addr:$dst, VR256:$src)>;
317 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
318 "movaps\t{$src, $dst|$dst, $src}",
319 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
320 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
321 "movapd\t{$src, $dst|$dst, $src}",
322 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
323 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
324 "movups\t{$src, $dst|$dst, $src}",
325 [(store (v4f32 VR128:$src), addr:$dst)]>;
326 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
327 "movupd\t{$src, $dst|$dst, $src}",
328 [(store (v2f64 VR128:$src), addr:$dst)]>;
330 // Intrinsic forms of MOVUPS/D load and store
331 let isAsmParserOnly = 1 in {
332 let canFoldAsLoad = 1, isReMaterializable = 1 in
333 def VMOVUPSrm_Int : VPSI<0x10, MRMSrcMem, (outs VR128:$dst),
335 "movups\t{$src, $dst|$dst, $src}",
336 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>, VEX;
337 def VMOVUPDrm_Int : VPDI<0x10, MRMSrcMem, (outs VR128:$dst),
339 "movupd\t{$src, $dst|$dst, $src}",
340 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>, VEX;
341 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
342 (ins f128mem:$dst, VR128:$src),
343 "movups\t{$src, $dst|$dst, $src}",
344 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
345 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
346 (ins f128mem:$dst, VR128:$src),
347 "movupd\t{$src, $dst|$dst, $src}",
348 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
350 let canFoldAsLoad = 1, isReMaterializable = 1 in
351 def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
352 "movups\t{$src, $dst|$dst, $src}",
353 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
354 def MOVUPDrm_Int : PDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
355 "movupd\t{$src, $dst|$dst, $src}",
356 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
358 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
359 "movups\t{$src, $dst|$dst, $src}",
360 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
361 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
362 "movupd\t{$src, $dst|$dst, $src}",
363 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
365 // Move Low/High packed floating point values
366 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
367 PatFrag mov_frag, string base_opc,
369 def PSrm : PI<opc, MRMSrcMem,
370 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
371 !strconcat(base_opc, "s", asm_opr),
374 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
375 SSEPackedSingle>, TB;
377 def PDrm : PI<opc, MRMSrcMem,
378 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
379 !strconcat(base_opc, "d", asm_opr),
380 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
381 (scalar_to_vector (loadf64 addr:$src2)))))],
382 SSEPackedDouble>, TB, OpSize;
385 let isAsmParserOnly = 1, AddedComplexity = 20 in {
386 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
387 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
388 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
389 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
391 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
392 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
393 "\t{$src2, $dst|$dst, $src2}">;
394 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
395 "\t{$src2, $dst|$dst, $src2}">;
398 let isAsmParserOnly = 1 in {
399 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
400 "movlps\t{$src, $dst|$dst, $src}",
401 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
402 (iPTR 0))), addr:$dst)]>, VEX;
403 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movlpd\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract (v2f64 VR128:$src),
406 (iPTR 0))), addr:$dst)]>, VEX;
408 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movlps\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
411 (iPTR 0))), addr:$dst)]>;
412 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
413 "movlpd\t{$src, $dst|$dst, $src}",
414 [(store (f64 (vector_extract (v2f64 VR128:$src),
415 (iPTR 0))), addr:$dst)]>;
417 // v2f64 extract element 1 is always custom lowered to unpack high to low
418 // and extract element 0 so the non-store version isn't too horrible.
419 let isAsmParserOnly = 1 in {
420 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
421 "movhps\t{$src, $dst|$dst, $src}",
422 [(store (f64 (vector_extract
423 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
424 (undef)), (iPTR 0))), addr:$dst)]>,
426 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
427 "movhpd\t{$src, $dst|$dst, $src}",
428 [(store (f64 (vector_extract
429 (v2f64 (unpckh VR128:$src, (undef))),
430 (iPTR 0))), addr:$dst)]>,
433 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
434 "movhps\t{$src, $dst|$dst, $src}",
435 [(store (f64 (vector_extract
436 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
437 (undef)), (iPTR 0))), addr:$dst)]>;
438 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
439 "movhpd\t{$src, $dst|$dst, $src}",
440 [(store (f64 (vector_extract
441 (v2f64 (unpckh VR128:$src, (undef))),
442 (iPTR 0))), addr:$dst)]>;
444 let isAsmParserOnly = 1, AddedComplexity = 20 in {
445 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
446 (ins VR128:$src1, VR128:$src2),
447 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
449 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
451 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
452 (ins VR128:$src1, VR128:$src2),
453 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
455 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
458 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
459 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
460 (ins VR128:$src1, VR128:$src2),
461 "movlhps\t{$src2, $dst|$dst, $src2}",
463 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
464 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
465 (ins VR128:$src1, VR128:$src2),
466 "movhlps\t{$src2, $dst|$dst, $src2}",
468 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
471 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
472 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
473 let AddedComplexity = 20 in {
474 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
475 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
476 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
477 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
480 //===----------------------------------------------------------------------===//
481 // SSE 1 & 2 - Conversion Instructions
482 //===----------------------------------------------------------------------===//
484 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
485 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
487 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
488 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
489 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
490 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
493 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
494 X86MemOperand x86memop, string asm> {
495 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
497 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
501 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
502 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
503 string asm, Domain d> {
504 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
505 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
506 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
507 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
510 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
511 X86MemOperand x86memop, string asm> {
512 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
513 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
514 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
515 (ins DstRC:$src1, x86memop:$src),
516 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
519 let isAsmParserOnly = 1 in {
520 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
521 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
522 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
523 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
525 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
526 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
527 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
528 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
531 // The assembler can recognize rr 64-bit instructions by seeing a rxx
532 // register, but the same isn't true when only using memory operands,
533 // provide other assembly "l" and "q" forms to address this explicitly
534 // where appropriate to do so.
535 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
537 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
539 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
541 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
543 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
547 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
548 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
549 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
550 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
551 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
552 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
553 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
554 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
555 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
556 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
557 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
558 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
559 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
560 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
561 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
562 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
564 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
565 // and/or XMM operand(s).
567 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
568 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
570 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
571 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
572 [(set DstRC:$dst, (Int SrcRC:$src))]>;
573 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
574 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
575 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
578 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
579 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
580 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
581 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
583 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
584 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
585 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
586 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
587 (ins DstRC:$src1, x86memop:$src2),
589 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
590 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
591 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
594 let isAsmParserOnly = 1 in {
595 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
596 f32mem, load, "cvtss2si">, XS, VEX;
597 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
598 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
600 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
601 f128mem, load, "cvtsd2si">, XD, VEX;
602 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
603 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
606 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
607 // Get rid of this hack or rename the intrinsics, there are several
608 // intructions that only match with the intrinsic form, why create duplicates
609 // to let them be recognized by the assembler?
610 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
611 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
612 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
613 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
615 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
616 f32mem, load, "cvtss2si">, XS;
617 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
618 f32mem, load, "cvtss2si{q}">, XS, REX_W;
619 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
620 f128mem, load, "cvtsd2si{l}">, XD;
621 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
622 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
625 let isAsmParserOnly = 1 in {
626 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
627 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
628 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
629 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
631 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
632 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
633 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
634 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
638 let Constraints = "$src1 = $dst" in {
639 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
640 int_x86_sse_cvtsi2ss, i32mem, loadi32,
642 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
643 int_x86_sse_cvtsi642ss, i64mem, loadi64,
644 "cvtsi2ss{q}">, XS, REX_W;
645 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
646 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
648 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
649 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
650 "cvtsi2sd">, XD, REX_W;
655 // Aliases for intrinsics
656 let isAsmParserOnly = 1 in {
657 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
658 f32mem, load, "cvttss2si">, XS, VEX;
659 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
660 int_x86_sse_cvttss2si64, f32mem, load,
661 "cvttss2si">, XS, VEX, VEX_W;
662 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
663 f128mem, load, "cvttsd2si">, XD, VEX;
664 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
665 int_x86_sse2_cvttsd2si64, f128mem, load,
666 "cvttsd2si">, XD, VEX, VEX_W;
668 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
669 f32mem, load, "cvttss2si">, XS;
670 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
671 int_x86_sse_cvttss2si64, f32mem, load,
672 "cvttss2si{q}">, XS, REX_W;
673 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
674 f128mem, load, "cvttsd2si">, XD;
675 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
676 int_x86_sse2_cvttsd2si64, f128mem, load,
677 "cvttsd2si{q}">, XD, REX_W;
679 let isAsmParserOnly = 1, Pattern = []<dag> in {
680 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
681 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
682 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
683 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
685 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
686 "cvtdq2ps\t{$src, $dst|$dst, $src}",
687 SSEPackedSingle>, TB, VEX;
688 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
689 "cvtdq2ps\t{$src, $dst|$dst, $src}",
690 SSEPackedSingle>, TB, VEX;
692 let Pattern = []<dag> in {
693 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
694 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
695 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
696 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
697 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
698 "cvtdq2ps\t{$src, $dst|$dst, $src}",
699 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
704 // Convert scalar double to scalar single
705 let isAsmParserOnly = 1 in {
706 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
707 (ins FR64:$src1, FR64:$src2),
708 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
710 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
711 (ins FR64:$src1, f64mem:$src2),
712 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
713 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
715 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
718 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
719 "cvtsd2ss\t{$src, $dst|$dst, $src}",
720 [(set FR32:$dst, (fround FR64:$src))]>;
721 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
722 "cvtsd2ss\t{$src, $dst|$dst, $src}",
723 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
724 Requires<[HasSSE2, OptForSize]>;
726 let isAsmParserOnly = 1 in
727 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
728 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
730 let Constraints = "$src1 = $dst" in
731 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
732 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
734 // Convert scalar single to scalar double
735 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
736 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
737 (ins FR32:$src1, FR32:$src2),
738 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
739 []>, XS, Requires<[HasAVX]>, VEX_4V;
740 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
741 (ins FR32:$src1, f32mem:$src2),
742 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
743 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
745 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
748 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
749 "cvtss2sd\t{$src, $dst|$dst, $src}",
750 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
752 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
753 "cvtss2sd\t{$src, $dst|$dst, $src}",
754 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
755 Requires<[HasSSE2, OptForSize]>;
757 let isAsmParserOnly = 1 in {
758 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
759 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
760 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
761 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
762 VR128:$src2))]>, XS, VEX_4V,
764 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
765 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
766 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
767 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
768 (load addr:$src2)))]>, XS, VEX_4V,
771 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
772 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
773 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
774 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
775 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
778 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
779 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
780 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
781 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
782 (load addr:$src2)))]>, XS,
786 def : Pat<(extloadf32 addr:$src),
787 (CVTSS2SDrr (MOVSSrm addr:$src))>,
788 Requires<[HasSSE2, OptForSpeed]>;
790 // Convert doubleword to packed single/double fp
791 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
792 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
793 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
794 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
795 TB, VEX, Requires<[HasAVX]>;
796 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
797 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
798 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
799 (bitconvert (memopv2i64 addr:$src))))]>,
800 TB, VEX, Requires<[HasAVX]>;
802 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
803 "cvtdq2ps\t{$src, $dst|$dst, $src}",
804 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
805 TB, Requires<[HasSSE2]>;
806 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
807 "cvtdq2ps\t{$src, $dst|$dst, $src}",
808 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
809 (bitconvert (memopv2i64 addr:$src))))]>,
810 TB, Requires<[HasSSE2]>;
812 // FIXME: why the non-intrinsic version is described as SSE3?
813 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
814 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
815 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
816 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
817 XS, VEX, Requires<[HasAVX]>;
818 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
819 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
820 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
821 (bitconvert (memopv2i64 addr:$src))))]>,
822 XS, VEX, Requires<[HasAVX]>;
824 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
825 "cvtdq2pd\t{$src, $dst|$dst, $src}",
826 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
827 XS, Requires<[HasSSE2]>;
828 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
829 "cvtdq2pd\t{$src, $dst|$dst, $src}",
830 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
831 (bitconvert (memopv2i64 addr:$src))))]>,
832 XS, Requires<[HasSSE2]>;
835 // Convert packed single/double fp to doubleword
836 let isAsmParserOnly = 1 in {
837 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
838 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
839 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
840 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
841 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
842 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
843 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
844 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
846 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
847 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
848 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
849 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
851 let isAsmParserOnly = 1 in {
852 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
853 "cvtps2dq\t{$src, $dst|$dst, $src}",
854 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
856 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
858 "cvtps2dq\t{$src, $dst|$dst, $src}",
859 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
860 (memop addr:$src)))]>, VEX;
862 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
863 "cvtps2dq\t{$src, $dst|$dst, $src}",
864 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
865 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
866 "cvtps2dq\t{$src, $dst|$dst, $src}",
867 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
868 (memop addr:$src)))]>;
870 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
871 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
872 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
873 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
874 XD, VEX, Requires<[HasAVX]>;
875 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
876 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
877 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
878 (memop addr:$src)))]>,
879 XD, VEX, Requires<[HasAVX]>;
881 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
882 "cvtpd2dq\t{$src, $dst|$dst, $src}",
883 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
884 XD, Requires<[HasSSE2]>;
885 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
886 "cvtpd2dq\t{$src, $dst|$dst, $src}",
887 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
888 (memop addr:$src)))]>,
889 XD, Requires<[HasSSE2]>;
892 // Convert with truncation packed single/double fp to doubleword
893 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
894 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
895 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
896 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
897 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
898 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
899 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
900 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
901 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
903 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
904 "cvttps2dq\t{$src, $dst|$dst, $src}",
906 (int_x86_sse2_cvttps2dq VR128:$src))]>;
907 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
908 "cvttps2dq\t{$src, $dst|$dst, $src}",
910 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
913 let isAsmParserOnly = 1 in {
914 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
915 "vcvttps2dq\t{$src, $dst|$dst, $src}",
917 (int_x86_sse2_cvttps2dq VR128:$src))]>,
918 XS, VEX, Requires<[HasAVX]>;
919 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
920 "vcvttps2dq\t{$src, $dst|$dst, $src}",
921 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
922 (memop addr:$src)))]>,
923 XS, VEX, Requires<[HasAVX]>;
926 let isAsmParserOnly = 1 in {
927 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
929 "cvttpd2dq\t{$src, $dst|$dst, $src}",
930 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
932 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
934 "cvttpd2dq\t{$src, $dst|$dst, $src}",
935 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
936 (memop addr:$src)))]>, VEX;
938 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
939 "cvttpd2dq\t{$src, $dst|$dst, $src}",
940 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
941 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
942 "cvttpd2dq\t{$src, $dst|$dst, $src}",
943 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
944 (memop addr:$src)))]>;
946 let isAsmParserOnly = 1 in {
947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
948 // register, but the same isn't true when using memory operands instead.
949 // Provide other assembly rr and rm forms to address this explicitly.
950 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
952 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
953 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
963 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
964 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
965 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
968 // Convert packed single to packed double
969 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
970 // SSE2 instructions without OpSize prefix
971 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
972 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
973 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
974 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
975 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
976 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
977 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
978 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
980 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
981 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
982 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
983 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
985 let isAsmParserOnly = 1 in {
986 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
987 "vcvtps2pd\t{$src, $dst|$dst, $src}",
988 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
989 VEX, Requires<[HasAVX]>;
990 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
991 "vcvtps2pd\t{$src, $dst|$dst, $src}",
992 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
993 (load addr:$src)))]>,
994 VEX, Requires<[HasAVX]>;
996 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
997 "cvtps2pd\t{$src, $dst|$dst, $src}",
998 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
999 TB, Requires<[HasSSE2]>;
1000 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1001 "cvtps2pd\t{$src, $dst|$dst, $src}",
1002 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1003 (load addr:$src)))]>,
1004 TB, Requires<[HasSSE2]>;
1006 // Convert packed double to packed single
1007 let isAsmParserOnly = 1 in {
1008 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1009 // register, but the same isn't true when using memory operands instead.
1010 // Provide other assembly rr and rm forms to address this explicitly.
1011 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1012 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1013 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1014 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1017 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1018 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1019 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1020 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1023 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1024 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1025 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1026 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1028 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1029 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1030 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1031 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1034 let isAsmParserOnly = 1 in {
1035 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1036 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1037 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1038 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1040 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1041 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1042 (memop addr:$src)))]>;
1044 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1045 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1046 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1047 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1048 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1049 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1050 (memop addr:$src)))]>;
1052 // AVX 256-bit register conversion intrinsics
1053 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1054 // whenever possible to avoid declaring two versions of each one.
1055 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1056 (VCVTDQ2PSYrr VR256:$src)>;
1057 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1058 (VCVTDQ2PSYrm addr:$src)>;
1060 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1061 (VCVTPD2PSYrr VR256:$src)>;
1062 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1063 (VCVTPD2PSYrm addr:$src)>;
1065 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1066 (VCVTPS2DQYrr VR256:$src)>;
1067 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1068 (VCVTPS2DQYrm addr:$src)>;
1070 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1071 (VCVTPS2PDYrr VR128:$src)>;
1072 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1073 (VCVTPS2PDYrm addr:$src)>;
1075 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1076 (VCVTTPD2DQYrr VR256:$src)>;
1077 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1078 (VCVTTPD2DQYrm addr:$src)>;
1080 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1081 (VCVTTPS2DQYrr VR256:$src)>;
1082 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1083 (VCVTTPS2DQYrm addr:$src)>;
1085 //===----------------------------------------------------------------------===//
1086 // SSE 1 & 2 - Compare Instructions
1087 //===----------------------------------------------------------------------===//
1089 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1090 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1091 string asm, string asm_alt> {
1092 def rr : SIi8<0xC2, MRMSrcReg,
1093 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1096 def rm : SIi8<0xC2, MRMSrcMem,
1097 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1099 // Accept explicit immediate argument form instead of comparison code.
1100 let isAsmParserOnly = 1 in {
1101 def rr_alt : SIi8<0xC2, MRMSrcReg,
1102 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1105 def rm_alt : SIi8<0xC2, MRMSrcMem,
1106 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1111 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1112 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1113 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1114 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1116 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1117 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1118 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1122 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1123 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1124 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1125 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1126 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1127 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1128 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1131 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1132 Intrinsic Int, string asm> {
1133 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1134 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1135 [(set VR128:$dst, (Int VR128:$src1,
1136 VR128:$src, imm:$cc))]>;
1137 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1138 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1139 [(set VR128:$dst, (Int VR128:$src1,
1140 (load addr:$src), imm:$cc))]>;
1143 // Aliases to match intrinsics which expect XMM operand(s).
1144 let isAsmParserOnly = 1 in {
1145 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1146 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1148 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1149 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1152 let Constraints = "$src1 = $dst" in {
1153 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1154 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1155 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1156 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1160 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1161 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1162 ValueType vt, X86MemOperand x86memop,
1163 PatFrag ld_frag, string OpcodeStr, Domain d> {
1164 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1165 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1166 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1167 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1168 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1169 [(set EFLAGS, (OpNode (vt RC:$src1),
1170 (ld_frag addr:$src2)))], d>;
1173 let Defs = [EFLAGS] in {
1174 let isAsmParserOnly = 1 in {
1175 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1176 "ucomiss", SSEPackedSingle>, VEX;
1177 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1178 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1179 let Pattern = []<dag> in {
1180 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1181 "comiss", SSEPackedSingle>, VEX;
1182 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1183 "comisd", SSEPackedDouble>, OpSize, VEX;
1186 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1187 load, "ucomiss", SSEPackedSingle>, VEX;
1188 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1189 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1191 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1192 load, "comiss", SSEPackedSingle>, VEX;
1193 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1194 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1196 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1197 "ucomiss", SSEPackedSingle>, TB;
1198 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1199 "ucomisd", SSEPackedDouble>, TB, OpSize;
1201 let Pattern = []<dag> in {
1202 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1203 "comiss", SSEPackedSingle>, TB;
1204 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1205 "comisd", SSEPackedDouble>, TB, OpSize;
1208 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1209 load, "ucomiss", SSEPackedSingle>, TB;
1210 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1211 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1213 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1214 "comiss", SSEPackedSingle>, TB;
1215 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1216 "comisd", SSEPackedDouble>, TB, OpSize;
1217 } // Defs = [EFLAGS]
1219 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1220 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1221 Intrinsic Int, string asm, string asm_alt,
1223 def rri : PIi8<0xC2, MRMSrcReg,
1224 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1225 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1226 def rmi : PIi8<0xC2, MRMSrcMem,
1227 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1228 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1229 // Accept explicit immediate argument form instead of comparison code.
1230 let isAsmParserOnly = 1 in {
1231 def rri_alt : PIi8<0xC2, MRMSrcReg,
1232 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1234 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1235 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1240 let isAsmParserOnly = 1 in {
1241 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1242 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1243 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1244 SSEPackedSingle>, VEX_4V;
1245 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1246 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1247 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1248 SSEPackedDouble>, OpSize, VEX_4V;
1249 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1250 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1251 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1252 SSEPackedSingle>, VEX_4V;
1253 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1254 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1255 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1256 SSEPackedDouble>, OpSize, VEX_4V;
1258 let Constraints = "$src1 = $dst" in {
1259 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1260 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1261 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1262 SSEPackedSingle>, TB;
1263 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1264 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1265 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1266 SSEPackedDouble>, TB, OpSize;
1269 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1270 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1271 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1272 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1273 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1274 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1275 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1276 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1278 //===----------------------------------------------------------------------===//
1279 // SSE 1 & 2 - Shuffle Instructions
1280 //===----------------------------------------------------------------------===//
1282 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1283 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1284 ValueType vt, string asm, PatFrag mem_frag,
1285 Domain d, bit IsConvertibleToThreeAddress = 0> {
1286 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1287 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1288 [(set RC:$dst, (vt (shufp:$src3
1289 RC:$src1, (mem_frag addr:$src2))))], d>;
1290 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1291 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1292 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1294 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1297 let isAsmParserOnly = 1 in {
1298 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1299 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1300 memopv4f32, SSEPackedSingle>, VEX_4V;
1301 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1302 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1303 memopv8f32, SSEPackedSingle>, VEX_4V;
1304 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1305 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1306 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1307 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1308 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1309 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1312 let Constraints = "$src1 = $dst" in {
1313 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1314 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1315 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1317 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1318 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1319 memopv2f64, SSEPackedDouble>, TB, OpSize;
1322 //===----------------------------------------------------------------------===//
1323 // SSE 1 & 2 - Unpack Instructions
1324 //===----------------------------------------------------------------------===//
1326 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1327 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1328 PatFrag mem_frag, RegisterClass RC,
1329 X86MemOperand x86memop, string asm,
1331 def rr : PI<opc, MRMSrcReg,
1332 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1334 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1335 def rm : PI<opc, MRMSrcMem,
1336 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1338 (vt (OpNode RC:$src1,
1339 (mem_frag addr:$src2))))], d>;
1342 let AddedComplexity = 10 in {
1343 let isAsmParserOnly = 1 in {
1344 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1345 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1346 SSEPackedSingle>, VEX_4V;
1347 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1348 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1349 SSEPackedDouble>, OpSize, VEX_4V;
1350 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1351 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1352 SSEPackedSingle>, VEX_4V;
1353 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1354 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1355 SSEPackedDouble>, OpSize, VEX_4V;
1357 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1358 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1359 SSEPackedSingle>, VEX_4V;
1360 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1361 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1362 SSEPackedDouble>, OpSize, VEX_4V;
1363 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1364 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1365 SSEPackedSingle>, VEX_4V;
1366 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1367 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1368 SSEPackedDouble>, OpSize, VEX_4V;
1371 let Constraints = "$src1 = $dst" in {
1372 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1373 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1374 SSEPackedSingle>, TB;
1375 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1376 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1377 SSEPackedDouble>, TB, OpSize;
1378 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1379 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1380 SSEPackedSingle>, TB;
1381 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1382 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1383 SSEPackedDouble>, TB, OpSize;
1384 } // Constraints = "$src1 = $dst"
1385 } // AddedComplexity
1387 //===----------------------------------------------------------------------===//
1388 // SSE 1 & 2 - Extract Floating-Point Sign mask
1389 //===----------------------------------------------------------------------===//
1391 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1392 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1394 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1395 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1396 [(set GR32:$dst, (Int RC:$src))], d>;
1397 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1398 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1402 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1403 SSEPackedSingle>, TB;
1404 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1405 SSEPackedDouble>, TB, OpSize;
1407 let isAsmParserOnly = 1 in {
1408 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1409 "movmskps", SSEPackedSingle>, VEX;
1410 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1411 "movmskpd", SSEPackedDouble>, OpSize,
1413 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1414 "movmskps", SSEPackedSingle>, VEX;
1415 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1416 "movmskpd", SSEPackedDouble>, OpSize,
1420 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1421 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1422 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1423 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1425 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1426 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1427 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1428 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1432 //===----------------------------------------------------------------------===//
1433 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1434 //===----------------------------------------------------------------------===//
1436 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1437 // names that start with 'Fs'.
1439 // Alias instructions that map fld0 to pxor for sse.
1440 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1441 canFoldAsLoad = 1 in {
1442 // FIXME: Set encoding to pseudo!
1443 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1444 [(set FR32:$dst, fp32imm0)]>,
1445 Requires<[HasSSE1]>, TB, OpSize;
1446 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1447 [(set FR64:$dst, fpimm0)]>,
1448 Requires<[HasSSE2]>, TB, OpSize;
1449 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1450 [(set FR32:$dst, fp32imm0)]>,
1451 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1452 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1453 [(set FR64:$dst, fpimm0)]>,
1454 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1457 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1458 // bits are disregarded.
1459 let neverHasSideEffects = 1 in {
1460 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1461 "movaps\t{$src, $dst|$dst, $src}", []>;
1462 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1463 "movapd\t{$src, $dst|$dst, $src}", []>;
1466 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1467 // bits are disregarded.
1468 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1469 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1470 "movaps\t{$src, $dst|$dst, $src}",
1471 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1472 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1473 "movapd\t{$src, $dst|$dst, $src}",
1474 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1477 //===----------------------------------------------------------------------===//
1478 // SSE 1 & 2 - Logical Instructions
1479 //===----------------------------------------------------------------------===//
1481 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1483 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1485 let isAsmParserOnly = 1 in {
1486 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1487 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1489 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1490 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1493 let Constraints = "$src1 = $dst" in {
1494 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1495 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1497 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1498 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1502 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1503 let mayLoad = 0 in {
1504 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1505 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1506 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1509 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1510 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1512 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1514 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1515 SDNode OpNode, int HasPat = 0,
1516 list<list<dag>> Pattern = []> {
1517 let isAsmParserOnly = 1, Pattern = []<dag> in {
1518 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1519 !strconcat(OpcodeStr, "ps"), f128mem,
1520 !if(HasPat, Pattern[0], // rr
1521 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1523 !if(HasPat, Pattern[2], // rm
1524 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1525 (memopv2i64 addr:$src2)))]), 0>,
1528 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1529 !strconcat(OpcodeStr, "pd"), f128mem,
1530 !if(HasPat, Pattern[1], // rr
1531 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1534 !if(HasPat, Pattern[3], // rm
1535 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1536 (memopv2i64 addr:$src2)))]), 0>,
1539 let Constraints = "$src1 = $dst" in {
1540 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1541 !strconcat(OpcodeStr, "ps"), f128mem,
1542 !if(HasPat, Pattern[0], // rr
1543 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1545 !if(HasPat, Pattern[2], // rm
1546 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1547 (memopv2i64 addr:$src2)))])>, TB;
1549 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1550 !strconcat(OpcodeStr, "pd"), f128mem,
1551 !if(HasPat, Pattern[1], // rr
1552 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1555 !if(HasPat, Pattern[3], // rm
1556 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1557 (memopv2i64 addr:$src2)))])>,
1562 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1564 let isAsmParserOnly = 1 in {
1565 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1566 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1567 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1569 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1570 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1574 // AVX 256-bit packed logical ops forms
1575 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1576 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1577 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1578 let isCommutable = 0 in
1579 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1581 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1582 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1583 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1584 let isCommutable = 0 in
1585 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1587 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1588 (bc_v2i64 (v4i32 immAllOnesV))),
1591 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1592 (bc_v2i64 (v2f64 VR128:$src2))))],
1594 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1595 (bc_v2i64 (v4i32 immAllOnesV))),
1596 (memopv2i64 addr:$src2))))],
1598 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1599 (memopv2i64 addr:$src2)))]]>;
1601 //===----------------------------------------------------------------------===//
1602 // SSE 1 & 2 - Arithmetic Instructions
1603 //===----------------------------------------------------------------------===//
1605 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1608 /// In addition, we also have a special variant of the scalar form here to
1609 /// represent the associated intrinsic operation. This form is unlike the
1610 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1611 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1613 /// These three forms can each be reg+reg or reg+mem.
1616 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1618 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1620 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1621 OpNode, FR32, f32mem, Is2Addr>, XS;
1622 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1623 OpNode, FR64, f64mem, Is2Addr>, XD;
1626 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1628 let mayLoad = 0 in {
1629 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1630 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1631 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1632 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1636 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1638 let mayLoad = 0 in {
1639 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1640 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1641 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1642 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1646 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1648 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1649 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1650 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1651 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1654 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1656 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1657 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1658 SSEPackedSingle, Is2Addr>, TB;
1660 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1661 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1662 SSEPackedDouble, Is2Addr>, TB, OpSize;
1665 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1666 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1667 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1668 SSEPackedSingle, 0>, TB;
1670 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1671 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1672 SSEPackedDouble, 0>, TB, OpSize;
1675 // Binary Arithmetic instructions
1676 let isAsmParserOnly = 1 in {
1677 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1678 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1679 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1680 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1681 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1682 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1683 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1684 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1686 let isCommutable = 0 in {
1687 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1688 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1689 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1690 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1691 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1692 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1693 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1694 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1695 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1696 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1697 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1698 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1699 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1700 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1701 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1702 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1703 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1704 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1705 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1706 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1710 let Constraints = "$src1 = $dst" in {
1711 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1712 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1713 basic_sse12_fp_binop_s_int<0x58, "add">;
1714 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1715 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1716 basic_sse12_fp_binop_s_int<0x59, "mul">;
1718 let isCommutable = 0 in {
1719 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1720 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1721 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1722 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1723 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1724 basic_sse12_fp_binop_s_int<0x5E, "div">;
1725 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1726 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1727 basic_sse12_fp_binop_s_int<0x5F, "max">,
1728 basic_sse12_fp_binop_p_int<0x5F, "max">;
1729 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1730 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1731 basic_sse12_fp_binop_s_int<0x5D, "min">,
1732 basic_sse12_fp_binop_p_int<0x5D, "min">;
1737 /// In addition, we also have a special variant of the scalar form here to
1738 /// represent the associated intrinsic operation. This form is unlike the
1739 /// plain scalar form, in that it takes an entire vector (instead of a
1740 /// scalar) and leaves the top elements undefined.
1742 /// And, we have a special variant form for a full-vector intrinsic form.
1744 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1745 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1746 SDNode OpNode, Intrinsic F32Int> {
1747 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1748 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1749 [(set FR32:$dst, (OpNode FR32:$src))]>;
1750 // For scalar unary operations, fold a load into the operation
1751 // only in OptForSize mode. It eliminates an instruction, but it also
1752 // eliminates a whole-register clobber (the load), so it introduces a
1753 // partial register update condition.
1754 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1755 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1756 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1757 Requires<[HasSSE1, OptForSize]>;
1758 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1759 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1760 [(set VR128:$dst, (F32Int VR128:$src))]>;
1761 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1762 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1763 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1766 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1767 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1768 SDNode OpNode, Intrinsic F32Int> {
1769 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1770 !strconcat(OpcodeStr,
1771 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1772 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1773 !strconcat(OpcodeStr,
1774 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1775 []>, XS, Requires<[HasAVX, OptForSize]>;
1776 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1777 !strconcat(OpcodeStr,
1778 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1779 [(set VR128:$dst, (F32Int VR128:$src))]>;
1780 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1781 !strconcat(OpcodeStr,
1782 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1783 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1786 /// sse1_fp_unop_p - SSE1 unops in packed form.
1787 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1788 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1789 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1790 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1791 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1792 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1793 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1796 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1797 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1798 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1799 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1800 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1801 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1802 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1803 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1806 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1807 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1808 Intrinsic V4F32Int> {
1809 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1810 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1811 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1812 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1813 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1814 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1817 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1818 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1819 Intrinsic V4F32Int> {
1820 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1821 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1822 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1823 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1824 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1825 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1828 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1829 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1830 SDNode OpNode, Intrinsic F64Int> {
1831 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1832 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1833 [(set FR64:$dst, (OpNode FR64:$src))]>;
1834 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1835 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1836 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1837 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1838 Requires<[HasSSE2, OptForSize]>;
1839 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1840 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1841 [(set VR128:$dst, (F64Int VR128:$src))]>;
1842 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1843 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1844 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1847 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1848 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1849 SDNode OpNode, Intrinsic F64Int> {
1850 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1851 !strconcat(OpcodeStr,
1852 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1853 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1854 (ins FR64:$src1, f64mem:$src2),
1855 !strconcat(OpcodeStr,
1856 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1857 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1858 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1859 [(set VR128:$dst, (F64Int VR128:$src))]>;
1860 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1861 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1862 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1865 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1866 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1868 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1869 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1870 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1871 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1872 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1873 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1876 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1877 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1878 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1879 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1880 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1881 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1882 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1883 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1886 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1887 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1888 Intrinsic V2F64Int> {
1889 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1890 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1891 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1892 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1893 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1894 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1897 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1898 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1899 Intrinsic V2F64Int> {
1900 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1901 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1902 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1903 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1904 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1905 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1908 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1910 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1911 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1914 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1915 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1916 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1917 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1918 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1919 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1920 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1921 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1924 // Reciprocal approximations. Note that these typically require refinement
1925 // in order to obtain suitable precision.
1926 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1927 int_x86_sse_rsqrt_ss>, VEX_4V;
1928 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1929 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1930 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1931 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1933 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1935 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1936 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1937 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1938 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1942 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1943 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1944 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1945 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1946 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1947 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1949 // Reciprocal approximations. Note that these typically require refinement
1950 // in order to obtain suitable precision.
1951 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1952 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1953 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1954 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1955 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1956 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1958 // There is no f64 version of the reciprocal approximation instructions.
1960 //===----------------------------------------------------------------------===//
1961 // SSE 1 & 2 - Non-temporal stores
1962 //===----------------------------------------------------------------------===//
1964 let isAsmParserOnly = 1 in {
1965 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1966 (ins i128mem:$dst, VR128:$src),
1967 "movntps\t{$src, $dst|$dst, $src}",
1968 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1969 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1970 (ins i128mem:$dst, VR128:$src),
1971 "movntpd\t{$src, $dst|$dst, $src}",
1972 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1974 let ExeDomain = SSEPackedInt in
1975 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1976 (ins f128mem:$dst, VR128:$src),
1977 "movntdq\t{$src, $dst|$dst, $src}",
1978 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1980 let AddedComplexity = 400 in { // Prefer non-temporal versions
1981 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1982 (ins f128mem:$dst, VR128:$src),
1983 "movntps\t{$src, $dst|$dst, $src}",
1984 [(alignednontemporalstore (v4f32 VR128:$src),
1986 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1987 (ins f128mem:$dst, VR128:$src),
1988 "movntpd\t{$src, $dst|$dst, $src}",
1989 [(alignednontemporalstore (v2f64 VR128:$src),
1991 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1992 (ins f128mem:$dst, VR128:$src),
1993 "movntdq\t{$src, $dst|$dst, $src}",
1994 [(alignednontemporalstore (v2f64 VR128:$src),
1996 let ExeDomain = SSEPackedInt in
1997 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1998 (ins f128mem:$dst, VR128:$src),
1999 "movntdq\t{$src, $dst|$dst, $src}",
2000 [(alignednontemporalstore (v4f32 VR128:$src),
2003 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2004 (ins f256mem:$dst, VR256:$src),
2005 "movntps\t{$src, $dst|$dst, $src}",
2006 [(alignednontemporalstore (v8f32 VR256:$src),
2008 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2009 (ins f256mem:$dst, VR256:$src),
2010 "movntpd\t{$src, $dst|$dst, $src}",
2011 [(alignednontemporalstore (v4f64 VR256:$src),
2013 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2014 (ins f256mem:$dst, VR256:$src),
2015 "movntdq\t{$src, $dst|$dst, $src}",
2016 [(alignednontemporalstore (v4f64 VR256:$src),
2018 let ExeDomain = SSEPackedInt in
2019 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2020 (ins f256mem:$dst, VR256:$src),
2021 "movntdq\t{$src, $dst|$dst, $src}",
2022 [(alignednontemporalstore (v8f32 VR256:$src),
2027 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2028 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2029 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2030 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2031 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2032 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2034 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2035 "movntps\t{$src, $dst|$dst, $src}",
2036 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2037 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2038 "movntpd\t{$src, $dst|$dst, $src}",
2039 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2041 let ExeDomain = SSEPackedInt in
2042 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2043 "movntdq\t{$src, $dst|$dst, $src}",
2044 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2046 let AddedComplexity = 400 in { // Prefer non-temporal versions
2047 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2048 "movntps\t{$src, $dst|$dst, $src}",
2049 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2050 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2051 "movntpd\t{$src, $dst|$dst, $src}",
2052 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2054 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2055 "movntdq\t{$src, $dst|$dst, $src}",
2056 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2058 let ExeDomain = SSEPackedInt in
2059 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2060 "movntdq\t{$src, $dst|$dst, $src}",
2061 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2063 // There is no AVX form for instructions below this point
2064 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2065 "movnti\t{$src, $dst|$dst, $src}",
2066 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2067 TB, Requires<[HasSSE2]>;
2069 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2070 "movnti\t{$src, $dst|$dst, $src}",
2071 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2072 TB, Requires<[HasSSE2]>;
2075 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2076 "movnti\t{$src, $dst|$dst, $src}",
2077 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2078 TB, Requires<[HasSSE2]>;
2080 //===----------------------------------------------------------------------===//
2081 // SSE 1 & 2 - Misc Instructions (No AVX form)
2082 //===----------------------------------------------------------------------===//
2084 // Prefetch intrinsic.
2085 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2086 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2087 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2088 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2089 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2090 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2091 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2092 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2094 // Load, store, and memory fence
2095 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2096 TB, Requires<[HasSSE1]>;
2097 def : Pat<(X86SFence), (SFENCE)>;
2099 // Alias instructions that map zero vector to pxor / xorp* for sse.
2100 // We set canFoldAsLoad because this can be converted to a constant-pool
2101 // load of an all-zeros value if folding it would be beneficial.
2102 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2103 // JIT implementation, it does not expand the instructions below like
2104 // X86MCInstLower does.
2105 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2106 isCodeGenOnly = 1 in {
2107 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2108 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2109 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2110 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2111 let ExeDomain = SSEPackedInt in
2112 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2113 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2116 // The same as done above but for AVX. The 128-bit versions are the
2117 // same, but re-encoded. The 256-bit does not support PI version.
2118 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2119 // JIT implementatioan, it does not expand the instructions below like
2120 // X86MCInstLower does.
2121 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2122 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2123 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2124 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2125 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2126 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2127 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2128 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2129 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2130 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2131 let ExeDomain = SSEPackedInt in
2132 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2133 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2136 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2137 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2138 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2140 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2141 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2143 //===----------------------------------------------------------------------===//
2144 // SSE 1 & 2 - Load/Store XCSR register
2145 //===----------------------------------------------------------------------===//
2147 let isAsmParserOnly = 1 in {
2148 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2149 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2150 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2151 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2154 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2155 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2156 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2157 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2159 //===---------------------------------------------------------------------===//
2160 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2161 //===---------------------------------------------------------------------===//
2163 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2165 let isAsmParserOnly = 1 in {
2166 let neverHasSideEffects = 1 in {
2167 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2168 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2169 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2170 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2172 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2173 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2174 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2175 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2177 let canFoldAsLoad = 1, mayLoad = 1 in {
2178 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2179 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2180 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2181 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2182 let Predicates = [HasAVX] in {
2183 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2184 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2185 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2186 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2190 let mayStore = 1 in {
2191 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2192 (ins i128mem:$dst, VR128:$src),
2193 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2194 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2195 (ins i256mem:$dst, VR256:$src),
2196 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2197 let Predicates = [HasAVX] in {
2198 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2199 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2200 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2201 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2206 let neverHasSideEffects = 1 in
2207 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2208 "movdqa\t{$src, $dst|$dst, $src}", []>;
2210 let canFoldAsLoad = 1, mayLoad = 1 in {
2211 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2212 "movdqa\t{$src, $dst|$dst, $src}",
2213 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2214 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2215 "movdqu\t{$src, $dst|$dst, $src}",
2216 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2217 XS, Requires<[HasSSE2]>;
2220 let mayStore = 1 in {
2221 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2222 "movdqa\t{$src, $dst|$dst, $src}",
2223 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2224 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2225 "movdqu\t{$src, $dst|$dst, $src}",
2226 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2227 XS, Requires<[HasSSE2]>;
2230 // Intrinsic forms of MOVDQU load and store
2231 let isAsmParserOnly = 1 in {
2232 let canFoldAsLoad = 1 in
2233 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2234 "vmovdqu\t{$src, $dst|$dst, $src}",
2235 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2236 XS, VEX, Requires<[HasAVX]>;
2237 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2238 "vmovdqu\t{$src, $dst|$dst, $src}",
2239 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2240 XS, VEX, Requires<[HasAVX]>;
2243 let canFoldAsLoad = 1 in
2244 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2245 "movdqu\t{$src, $dst|$dst, $src}",
2246 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2247 XS, Requires<[HasSSE2]>;
2248 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2249 "movdqu\t{$src, $dst|$dst, $src}",
2250 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2251 XS, Requires<[HasSSE2]>;
2253 } // ExeDomain = SSEPackedInt
2255 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2256 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2257 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2259 //===---------------------------------------------------------------------===//
2260 // SSE2 - Packed Integer Arithmetic Instructions
2261 //===---------------------------------------------------------------------===//
2263 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2265 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2266 bit IsCommutable = 0, bit Is2Addr = 1> {
2267 let isCommutable = IsCommutable in
2268 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2269 (ins VR128:$src1, VR128:$src2),
2271 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2272 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2273 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2274 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2275 (ins VR128:$src1, i128mem:$src2),
2277 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2278 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2279 [(set VR128:$dst, (IntId VR128:$src1,
2280 (bitconvert (memopv2i64 addr:$src2))))]>;
2283 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2284 string OpcodeStr, Intrinsic IntId,
2285 Intrinsic IntId2, bit Is2Addr = 1> {
2286 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2287 (ins VR128:$src1, VR128:$src2),
2289 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2290 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2291 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2292 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2293 (ins VR128:$src1, i128mem:$src2),
2295 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2296 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2297 [(set VR128:$dst, (IntId VR128:$src1,
2298 (bitconvert (memopv2i64 addr:$src2))))]>;
2299 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2300 (ins VR128:$src1, i32i8imm:$src2),
2302 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2303 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2304 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2307 /// PDI_binop_rm - Simple SSE2 binary operator.
2308 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2309 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2310 let isCommutable = IsCommutable in
2311 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2312 (ins VR128:$src1, VR128:$src2),
2314 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2316 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2317 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2318 (ins VR128:$src1, i128mem:$src2),
2320 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2321 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2322 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2323 (bitconvert (memopv2i64 addr:$src2)))))]>;
2326 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2328 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2329 /// to collapse (bitconvert VT to VT) into its operand.
2331 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2332 bit IsCommutable = 0, bit Is2Addr = 1> {
2333 let isCommutable = IsCommutable in
2334 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2335 (ins VR128:$src1, VR128:$src2),
2337 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2339 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2340 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2341 (ins VR128:$src1, i128mem:$src2),
2343 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2345 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2348 } // ExeDomain = SSEPackedInt
2350 // 128-bit Integer Arithmetic
2352 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2353 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2354 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2355 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2356 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2357 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2358 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2359 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2360 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2361 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2364 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2366 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2368 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2370 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2372 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2374 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2376 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2378 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2380 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2382 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2384 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2386 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2388 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2390 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2392 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2394 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2396 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2398 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2400 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2404 let Constraints = "$src1 = $dst" in {
2405 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2406 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2407 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2408 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2409 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2410 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2411 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2412 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2413 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2416 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2417 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2418 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2419 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2420 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2421 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2422 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2423 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2424 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2425 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2426 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2427 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2428 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2429 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2430 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2431 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2432 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2433 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2434 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2436 } // Constraints = "$src1 = $dst"
2438 //===---------------------------------------------------------------------===//
2439 // SSE2 - Packed Integer Logical Instructions
2440 //===---------------------------------------------------------------------===//
2442 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2443 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2444 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2446 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2447 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2449 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2450 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2453 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2454 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2456 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2457 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2459 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2460 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2463 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2464 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2466 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2467 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2470 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2471 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2472 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2474 let ExeDomain = SSEPackedInt in {
2475 let neverHasSideEffects = 1 in {
2476 // 128-bit logical shifts.
2477 def VPSLLDQri : PDIi8<0x73, MRM7r,
2478 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2479 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2481 def VPSRLDQri : PDIi8<0x73, MRM3r,
2482 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2483 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2485 // PSRADQri doesn't exist in SSE[1-3].
2487 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2488 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2489 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2490 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2491 VR128:$src2)))]>, VEX_4V;
2493 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2494 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2495 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2496 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2497 (memopv2i64 addr:$src2))))]>,
2502 let Constraints = "$src1 = $dst" in {
2503 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2504 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2505 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2506 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2507 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2508 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2510 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2511 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2512 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2513 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2514 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2515 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2517 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2518 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2519 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2520 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2522 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2523 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2524 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2526 let ExeDomain = SSEPackedInt in {
2527 let neverHasSideEffects = 1 in {
2528 // 128-bit logical shifts.
2529 def PSLLDQri : PDIi8<0x73, MRM7r,
2530 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2531 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2532 def PSRLDQri : PDIi8<0x73, MRM3r,
2533 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2534 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2535 // PSRADQri doesn't exist in SSE[1-3].
2537 def PANDNrr : PDI<0xDF, MRMSrcReg,
2538 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2539 "pandn\t{$src2, $dst|$dst, $src2}",
2540 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2543 def PANDNrm : PDI<0xDF, MRMSrcMem,
2544 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2545 "pandn\t{$src2, $dst|$dst, $src2}",
2546 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2547 (memopv2i64 addr:$src2))))]>;
2549 } // Constraints = "$src1 = $dst"
2551 let Predicates = [HasAVX] in {
2552 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2553 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2554 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2555 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2556 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2557 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2558 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2559 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2560 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2561 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2563 // Shift up / down and insert zero's.
2564 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2565 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2566 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2567 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2570 let Predicates = [HasSSE2] in {
2571 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2572 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2573 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2574 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2575 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2576 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2577 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2578 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2579 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2580 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2582 // Shift up / down and insert zero's.
2583 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2584 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2585 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2586 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2589 //===---------------------------------------------------------------------===//
2590 // SSE2 - Packed Integer Comparison Instructions
2591 //===---------------------------------------------------------------------===//
2593 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2594 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2596 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2598 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2600 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2602 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2604 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2608 let Constraints = "$src1 = $dst" in {
2609 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2610 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2611 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2612 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2613 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2614 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2615 } // Constraints = "$src1 = $dst"
2617 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2618 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2619 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2620 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2621 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2622 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2623 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2624 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2625 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2626 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2627 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2628 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2630 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2631 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2632 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2633 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2634 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2635 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2636 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2637 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2638 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2639 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2640 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2641 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2643 //===---------------------------------------------------------------------===//
2644 // SSE2 - Packed Integer Pack Instructions
2645 //===---------------------------------------------------------------------===//
2647 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2648 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2650 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2652 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2656 let Constraints = "$src1 = $dst" in {
2657 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2658 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2659 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2660 } // Constraints = "$src1 = $dst"
2662 //===---------------------------------------------------------------------===//
2663 // SSE2 - Packed Integer Shuffle Instructions
2664 //===---------------------------------------------------------------------===//
2666 let ExeDomain = SSEPackedInt in {
2667 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2669 def ri : Ii8<0x70, MRMSrcReg,
2670 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2671 !strconcat(OpcodeStr,
2672 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2673 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2675 def mi : Ii8<0x70, MRMSrcMem,
2676 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2677 !strconcat(OpcodeStr,
2678 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2679 [(set VR128:$dst, (vt (pshuf_frag:$src2
2680 (bc_frag (memopv2i64 addr:$src1)),
2683 } // ExeDomain = SSEPackedInt
2685 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2686 let AddedComplexity = 5 in
2687 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2690 // SSE2 with ImmT == Imm8 and XS prefix.
2691 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2694 // SSE2 with ImmT == Imm8 and XD prefix.
2695 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2699 let Predicates = [HasSSE2] in {
2700 let AddedComplexity = 5 in
2701 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2703 // SSE2 with ImmT == Imm8 and XS prefix.
2704 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2706 // SSE2 with ImmT == Imm8 and XD prefix.
2707 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2710 //===---------------------------------------------------------------------===//
2711 // SSE2 - Packed Integer Unpack Instructions
2712 //===---------------------------------------------------------------------===//
2714 let ExeDomain = SSEPackedInt in {
2715 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2716 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2717 def rr : PDI<opc, MRMSrcReg,
2718 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2720 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2721 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2722 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2723 def rm : PDI<opc, MRMSrcMem,
2724 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2726 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2727 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2728 [(set VR128:$dst, (unp_frag VR128:$src1,
2729 (bc_frag (memopv2i64
2733 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2734 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2736 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2738 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2741 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2742 /// knew to collapse (bitconvert VT to VT) into its operand.
2743 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2744 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2745 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2747 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2748 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2749 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2750 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2752 (v2i64 (unpckl VR128:$src1,
2753 (memopv2i64 addr:$src2))))]>, VEX_4V;
2755 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2757 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2759 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2762 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2763 /// knew to collapse (bitconvert VT to VT) into its operand.
2764 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2765 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2766 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2768 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2769 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2770 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2771 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2773 (v2i64 (unpckh VR128:$src1,
2774 (memopv2i64 addr:$src2))))]>, VEX_4V;
2777 let Constraints = "$src1 = $dst" in {
2778 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2779 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2780 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2782 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2783 /// knew to collapse (bitconvert VT to VT) into its operand.
2784 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2785 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2786 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2788 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2789 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2790 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2791 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2793 (v2i64 (unpckl VR128:$src1,
2794 (memopv2i64 addr:$src2))))]>;
2796 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2797 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2798 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2800 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2801 /// knew to collapse (bitconvert VT to VT) into its operand.
2802 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2803 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2804 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2806 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2807 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2808 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2809 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2811 (v2i64 (unpckh VR128:$src1,
2812 (memopv2i64 addr:$src2))))]>;
2815 } // ExeDomain = SSEPackedInt
2817 //===---------------------------------------------------------------------===//
2818 // SSE2 - Packed Integer Extract and Insert
2819 //===---------------------------------------------------------------------===//
2821 let ExeDomain = SSEPackedInt in {
2822 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2823 def rri : Ii8<0xC4, MRMSrcReg,
2824 (outs VR128:$dst), (ins VR128:$src1,
2825 GR32:$src2, i32i8imm:$src3),
2827 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2828 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2830 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2831 def rmi : Ii8<0xC4, MRMSrcMem,
2832 (outs VR128:$dst), (ins VR128:$src1,
2833 i16mem:$src2, i32i8imm:$src3),
2835 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2836 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2838 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2843 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2844 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2845 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2846 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2847 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2848 imm:$src2))]>, OpSize, VEX;
2849 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2850 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2851 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2852 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2856 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2857 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2858 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2859 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2860 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2861 []>, OpSize, VEX_4V;
2864 let Constraints = "$src1 = $dst" in
2865 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2867 } // ExeDomain = SSEPackedInt
2869 //===---------------------------------------------------------------------===//
2870 // SSE2 - Packed Mask Creation
2871 //===---------------------------------------------------------------------===//
2873 let ExeDomain = SSEPackedInt in {
2875 let isAsmParserOnly = 1 in {
2876 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2877 "pmovmskb\t{$src, $dst|$dst, $src}",
2878 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2879 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2880 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2882 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2883 "pmovmskb\t{$src, $dst|$dst, $src}",
2884 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2886 } // ExeDomain = SSEPackedInt
2888 //===---------------------------------------------------------------------===//
2889 // SSE2 - Conditional Store
2890 //===---------------------------------------------------------------------===//
2892 let ExeDomain = SSEPackedInt in {
2894 let isAsmParserOnly = 1 in {
2896 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2897 (ins VR128:$src, VR128:$mask),
2898 "maskmovdqu\t{$mask, $src|$src, $mask}",
2899 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2901 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2902 (ins VR128:$src, VR128:$mask),
2903 "maskmovdqu\t{$mask, $src|$src, $mask}",
2904 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2908 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2909 "maskmovdqu\t{$mask, $src|$src, $mask}",
2910 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2912 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2913 "maskmovdqu\t{$mask, $src|$src, $mask}",
2914 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2916 } // ExeDomain = SSEPackedInt
2918 //===---------------------------------------------------------------------===//
2919 // SSE2 - Move Doubleword
2920 //===---------------------------------------------------------------------===//
2922 // Move Int Doubleword to Packed Double Int
2923 let isAsmParserOnly = 1 in {
2924 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2925 "movd\t{$src, $dst|$dst, $src}",
2927 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2928 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2929 "movd\t{$src, $dst|$dst, $src}",
2931 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2934 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2935 "movd\t{$src, $dst|$dst, $src}",
2937 (v4i32 (scalar_to_vector GR32:$src)))]>;
2938 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2939 "movd\t{$src, $dst|$dst, $src}",
2941 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2942 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2943 "mov{d|q}\t{$src, $dst|$dst, $src}",
2945 (v2i64 (scalar_to_vector GR64:$src)))]>;
2946 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2947 "mov{d|q}\t{$src, $dst|$dst, $src}",
2948 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2951 // Move Int Doubleword to Single Scalar
2952 let isAsmParserOnly = 1 in {
2953 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2954 "movd\t{$src, $dst|$dst, $src}",
2955 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2957 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2958 "movd\t{$src, $dst|$dst, $src}",
2959 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2962 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2963 "movd\t{$src, $dst|$dst, $src}",
2964 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2966 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2967 "movd\t{$src, $dst|$dst, $src}",
2968 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2970 // Move Packed Doubleword Int to Packed Double Int
2971 let isAsmParserOnly = 1 in {
2972 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2973 "movd\t{$src, $dst|$dst, $src}",
2974 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2976 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2977 (ins i32mem:$dst, VR128:$src),
2978 "movd\t{$src, $dst|$dst, $src}",
2979 [(store (i32 (vector_extract (v4i32 VR128:$src),
2980 (iPTR 0))), addr:$dst)]>, VEX;
2982 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2983 "movd\t{$src, $dst|$dst, $src}",
2984 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2986 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2987 "movd\t{$src, $dst|$dst, $src}",
2988 [(store (i32 (vector_extract (v4i32 VR128:$src),
2989 (iPTR 0))), addr:$dst)]>;
2991 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2992 "mov{d|q}\t{$src, $dst|$dst, $src}",
2993 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2995 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2996 "movq\t{$src, $dst|$dst, $src}",
2997 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2999 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
3000 "mov{d|q}\t{$src, $dst|$dst, $src}",
3001 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3002 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3003 "movq\t{$src, $dst|$dst, $src}",
3004 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3006 // Move Scalar Single to Double Int
3007 let isAsmParserOnly = 1 in {
3008 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3009 "movd\t{$src, $dst|$dst, $src}",
3010 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3011 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3012 "movd\t{$src, $dst|$dst, $src}",
3013 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3015 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3016 "movd\t{$src, $dst|$dst, $src}",
3017 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3018 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3019 "movd\t{$src, $dst|$dst, $src}",
3020 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3022 // movd / movq to XMM register zero-extends
3023 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3024 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3025 "movd\t{$src, $dst|$dst, $src}",
3026 [(set VR128:$dst, (v4i32 (X86vzmovl
3027 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3029 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3030 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3031 [(set VR128:$dst, (v2i64 (X86vzmovl
3032 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3035 let AddedComplexity = 15 in {
3036 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3037 "movd\t{$src, $dst|$dst, $src}",
3038 [(set VR128:$dst, (v4i32 (X86vzmovl
3039 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3040 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3041 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3042 [(set VR128:$dst, (v2i64 (X86vzmovl
3043 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3046 let AddedComplexity = 20 in {
3047 let isAsmParserOnly = 1 in
3048 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3049 "movd\t{$src, $dst|$dst, $src}",
3051 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3052 (loadi32 addr:$src))))))]>,
3054 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3055 "movd\t{$src, $dst|$dst, $src}",
3057 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3058 (loadi32 addr:$src))))))]>;
3060 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3061 (MOVZDI2PDIrm addr:$src)>;
3062 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3063 (MOVZDI2PDIrm addr:$src)>;
3064 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3065 (MOVZDI2PDIrm addr:$src)>;
3068 //===---------------------------------------------------------------------===//
3069 // SSE2 - Move Quadword
3070 //===---------------------------------------------------------------------===//
3072 // Move Quadword Int to Packed Quadword Int
3073 let isAsmParserOnly = 1 in
3074 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3075 "vmovq\t{$src, $dst|$dst, $src}",
3077 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3078 VEX, Requires<[HasAVX]>;
3079 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3080 "movq\t{$src, $dst|$dst, $src}",
3082 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3083 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3085 // Move Packed Quadword Int to Quadword Int
3086 let isAsmParserOnly = 1 in
3087 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3088 "movq\t{$src, $dst|$dst, $src}",
3089 [(store (i64 (vector_extract (v2i64 VR128:$src),
3090 (iPTR 0))), addr:$dst)]>, VEX;
3091 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3092 "movq\t{$src, $dst|$dst, $src}",
3093 [(store (i64 (vector_extract (v2i64 VR128:$src),
3094 (iPTR 0))), addr:$dst)]>;
3096 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3097 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3099 // Store / copy lower 64-bits of a XMM register.
3100 let isAsmParserOnly = 1 in
3101 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3102 "movq\t{$src, $dst|$dst, $src}",
3103 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3104 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3105 "movq\t{$src, $dst|$dst, $src}",
3106 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3108 let AddedComplexity = 20, isAsmParserOnly = 1 in
3109 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3110 "vmovq\t{$src, $dst|$dst, $src}",
3112 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3113 (loadi64 addr:$src))))))]>,
3114 XS, VEX, Requires<[HasAVX]>;
3116 let AddedComplexity = 20 in {
3117 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3118 "movq\t{$src, $dst|$dst, $src}",
3120 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3121 (loadi64 addr:$src))))))]>,
3122 XS, Requires<[HasSSE2]>;
3124 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3125 (MOVZQI2PQIrm addr:$src)>;
3126 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3127 (MOVZQI2PQIrm addr:$src)>;
3128 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3131 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3132 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3133 let isAsmParserOnly = 1, AddedComplexity = 15 in
3134 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3135 "vmovq\t{$src, $dst|$dst, $src}",
3136 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3137 XS, VEX, Requires<[HasAVX]>;
3138 let AddedComplexity = 15 in
3139 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3140 "movq\t{$src, $dst|$dst, $src}",
3141 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3142 XS, Requires<[HasSSE2]>;
3144 let AddedComplexity = 20, isAsmParserOnly = 1 in
3145 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3146 "vmovq\t{$src, $dst|$dst, $src}",
3147 [(set VR128:$dst, (v2i64 (X86vzmovl
3148 (loadv2i64 addr:$src))))]>,
3149 XS, VEX, Requires<[HasAVX]>;
3150 let AddedComplexity = 20 in {
3151 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3152 "movq\t{$src, $dst|$dst, $src}",
3153 [(set VR128:$dst, (v2i64 (X86vzmovl
3154 (loadv2i64 addr:$src))))]>,
3155 XS, Requires<[HasSSE2]>;
3157 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3158 (MOVZPQILo2PQIrm addr:$src)>;
3161 // Instructions to match in the assembler
3162 let isAsmParserOnly = 1 in {
3163 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3164 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3165 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3166 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3167 // Recognize "movd" with GR64 destination, but encode as a "movq"
3168 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3169 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3172 // Instructions for the disassembler
3173 // xr = XMM register
3176 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3177 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3178 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3179 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3180 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3182 //===---------------------------------------------------------------------===//
3183 // SSE2 - Misc Instructions
3184 //===---------------------------------------------------------------------===//
3187 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3188 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3189 TB, Requires<[HasSSE2]>;
3191 // Load, store, and memory fence
3192 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3193 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3194 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3195 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3196 def : Pat<(X86LFence), (LFENCE)>;
3197 def : Pat<(X86MFence), (MFENCE)>;
3200 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3201 // was introduced with SSE2, it's backward compatible.
3202 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3204 // Alias instructions that map zero vector to pxor / xorp* for sse.
3205 // We set canFoldAsLoad because this can be converted to a constant-pool
3206 // load of an all-ones value if folding it would be beneficial.
3207 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3208 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3209 // FIXME: Change encoding to pseudo.
3210 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3211 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3213 //===---------------------------------------------------------------------===//
3214 // SSE3 - Conversion Instructions
3215 //===---------------------------------------------------------------------===//
3217 // Convert Packed Double FP to Packed DW Integers
3218 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3219 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3220 // register, but the same isn't true when using memory operands instead.
3221 // Provide other assembly rr and rm forms to address this explicitly.
3222 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3223 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3224 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3225 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3228 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3229 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3230 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3231 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3234 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3235 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3236 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3237 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3240 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3241 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3242 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3243 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3245 // Convert Packed DW Integers to Packed Double FP
3246 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3247 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3248 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3249 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3250 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3251 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3252 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3253 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3254 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3257 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3258 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3259 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3260 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3262 // AVX 256-bit register conversion intrinsics
3263 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3264 (VCVTDQ2PDYrr VR128:$src)>;
3265 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3266 (VCVTDQ2PDYrm addr:$src)>;
3268 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3269 (VCVTPD2DQYrr VR256:$src)>;
3270 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3271 (VCVTPD2DQYrm addr:$src)>;
3273 //===---------------------------------------------------------------------===//
3274 // SSE3 - Move Instructions
3275 //===---------------------------------------------------------------------===//
3277 // Replicate Single FP
3278 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3279 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3280 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3281 [(set VR128:$dst, (v4f32 (rep_frag
3282 VR128:$src, (undef))))]>;
3283 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3284 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3285 [(set VR128:$dst, (rep_frag
3286 (memopv4f32 addr:$src), (undef)))]>;
3289 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3291 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3292 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3293 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3294 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3297 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3298 // FIXME: Merge above classes when we have patterns for the ymm version
3299 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3300 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3301 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3302 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3304 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3305 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3307 // Replicate Double FP
3308 multiclass sse3_replicate_dfp<string OpcodeStr> {
3309 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3310 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3311 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3312 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3313 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3315 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3319 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3320 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3321 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3323 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3324 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3328 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3329 // FIXME: Merge above classes when we have patterns for the ymm version
3330 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3331 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3333 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3335 // Move Unaligned Integer
3336 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3337 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3338 "vlddqu\t{$src, $dst|$dst, $src}",
3339 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3340 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3341 "vlddqu\t{$src, $dst|$dst, $src}",
3342 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3344 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3345 "lddqu\t{$src, $dst|$dst, $src}",
3346 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3348 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3350 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3352 // Several Move patterns
3353 let AddedComplexity = 5 in {
3354 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3355 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3356 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3357 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3358 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3359 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3360 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3361 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3364 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3365 let AddedComplexity = 15 in
3366 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3367 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3368 let AddedComplexity = 20 in
3369 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3370 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3372 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3373 let AddedComplexity = 15 in
3374 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3375 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3376 let AddedComplexity = 20 in
3377 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3378 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3380 //===---------------------------------------------------------------------===//
3381 // SSE3 - Arithmetic
3382 //===---------------------------------------------------------------------===//
3384 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3385 X86MemOperand x86memop, bit Is2Addr = 1> {
3386 def rr : I<0xD0, MRMSrcReg,
3387 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3389 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3390 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3391 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3392 def rm : I<0xD0, MRMSrcMem,
3393 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3395 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3396 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3397 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3400 let isAsmParserOnly = 1, Predicates = [HasAVX],
3401 ExeDomain = SSEPackedDouble in {
3402 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3403 f128mem, 0>, XD, VEX_4V;
3404 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3405 f128mem, 0>, OpSize, VEX_4V;
3406 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3407 f256mem, 0>, XD, VEX_4V;
3408 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3409 f256mem, 0>, OpSize, VEX_4V;
3411 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3412 ExeDomain = SSEPackedDouble in {
3413 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3415 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3416 f128mem>, TB, OpSize;
3419 //===---------------------------------------------------------------------===//
3420 // SSE3 Instructions
3421 //===---------------------------------------------------------------------===//
3424 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3425 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3426 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3428 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3429 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3430 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3432 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3434 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3435 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3436 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3438 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3439 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3440 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3444 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3446 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3448 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3449 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3450 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3453 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3454 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3455 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3456 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3457 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3458 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3459 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3460 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3461 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3462 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3463 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3464 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3465 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3466 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3467 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3468 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3469 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3472 let Constraints = "$src1 = $dst" in {
3473 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3474 int_x86_sse3_hadd_ps>;
3475 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3476 int_x86_sse3_hadd_pd>;
3477 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3478 int_x86_sse3_hsub_ps>;
3479 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3480 int_x86_sse3_hsub_pd>;
3483 //===---------------------------------------------------------------------===//
3484 // SSSE3 - Packed Absolute Instructions
3485 //===---------------------------------------------------------------------===//
3488 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3489 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3490 PatFrag mem_frag128, Intrinsic IntId128> {
3491 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3493 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3494 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3497 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3499 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3502 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3505 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3506 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3507 int_x86_ssse3_pabs_b_128>, VEX;
3508 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3509 int_x86_ssse3_pabs_w_128>, VEX;
3510 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3511 int_x86_ssse3_pabs_d_128>, VEX;
3514 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3515 int_x86_ssse3_pabs_b_128>;
3516 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3517 int_x86_ssse3_pabs_w_128>;
3518 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3519 int_x86_ssse3_pabs_d_128>;
3521 //===---------------------------------------------------------------------===//
3522 // SSSE3 - Packed Binary Operator Instructions
3523 //===---------------------------------------------------------------------===//
3525 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3526 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3527 PatFrag mem_frag128, Intrinsic IntId128,
3529 let isCommutable = 1 in
3530 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3531 (ins VR128:$src1, VR128:$src2),
3533 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3534 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3535 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3537 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3538 (ins VR128:$src1, i128mem:$src2),
3540 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3541 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3543 (IntId128 VR128:$src1,
3544 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3547 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3548 let isCommutable = 0 in {
3549 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3550 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3551 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3552 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3553 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3554 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3555 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3556 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3557 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3558 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3559 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3560 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3561 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3562 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3563 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3564 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3565 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3566 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3567 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3568 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3569 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3570 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3572 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3573 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3576 // None of these have i8 immediate fields.
3577 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3578 let isCommutable = 0 in {
3579 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3580 int_x86_ssse3_phadd_w_128>;
3581 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3582 int_x86_ssse3_phadd_d_128>;
3583 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3584 int_x86_ssse3_phadd_sw_128>;
3585 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3586 int_x86_ssse3_phsub_w_128>;
3587 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3588 int_x86_ssse3_phsub_d_128>;
3589 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3590 int_x86_ssse3_phsub_sw_128>;
3591 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3592 int_x86_ssse3_pmadd_ub_sw_128>;
3593 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3594 int_x86_ssse3_pshuf_b_128>;
3595 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3596 int_x86_ssse3_psign_b_128>;
3597 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3598 int_x86_ssse3_psign_w_128>;
3599 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3600 int_x86_ssse3_psign_d_128>;
3602 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3603 int_x86_ssse3_pmul_hr_sw_128>;
3606 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3607 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3608 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3609 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3611 //===---------------------------------------------------------------------===//
3612 // SSSE3 - Packed Align Instruction Patterns
3613 //===---------------------------------------------------------------------===//
3615 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3616 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3617 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3619 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3621 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3623 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3624 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3626 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3628 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3632 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3633 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3634 let Constraints = "$src1 = $dst" in
3635 defm PALIGN : ssse3_palign<"palignr">;
3637 let AddedComplexity = 5 in {
3638 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3639 (PALIGNR128rr VR128:$src2, VR128:$src1,
3640 (SHUFFLE_get_palign_imm VR128:$src3))>,
3641 Requires<[HasSSSE3]>;
3642 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3643 (PALIGNR128rr VR128:$src2, VR128:$src1,
3644 (SHUFFLE_get_palign_imm VR128:$src3))>,
3645 Requires<[HasSSSE3]>;
3646 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3647 (PALIGNR128rr VR128:$src2, VR128:$src1,
3648 (SHUFFLE_get_palign_imm VR128:$src3))>,
3649 Requires<[HasSSSE3]>;
3650 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3651 (PALIGNR128rr VR128:$src2, VR128:$src1,
3652 (SHUFFLE_get_palign_imm VR128:$src3))>,
3653 Requires<[HasSSSE3]>;
3656 //===---------------------------------------------------------------------===//
3657 // SSSE3 Misc Instructions
3658 //===---------------------------------------------------------------------===//
3660 // Thread synchronization
3661 let usesCustomInserter = 1 in {
3662 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3663 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3664 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3665 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3668 let Uses = [EAX, ECX, EDX] in
3669 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3670 Requires<[HasSSE3]>;
3671 let Uses = [ECX, EAX] in
3672 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3673 Requires<[HasSSE3]>;
3675 //===---------------------------------------------------------------------===//
3676 // Non-Instruction Patterns
3677 //===---------------------------------------------------------------------===//
3679 // extload f32 -> f64. This matches load+fextend because we have a hack in
3680 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3682 // Since these loads aren't folded into the fextend, we have to match it
3684 let Predicates = [HasSSE2] in
3685 def : Pat<(fextend (loadf32 addr:$src)),
3686 (CVTSS2SDrm addr:$src)>;
3689 let Predicates = [HasXMMInt] in {
3690 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3691 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3692 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3693 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3694 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3695 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3696 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3697 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3698 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3699 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3700 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3701 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3702 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3703 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3704 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3705 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3706 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3707 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3708 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3709 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3710 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3711 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3712 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3713 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3714 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3715 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3716 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3717 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3718 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3719 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3722 let Predicates = [HasAVX] in {
3723 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3726 // Move scalar to XMM zero-extended
3727 // movd to XMM register zero-extends
3728 let AddedComplexity = 15 in {
3729 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3730 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3731 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3732 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3733 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3734 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3735 (MOVSSrr (v4f32 (V_SET0PS)),
3736 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3737 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3738 (MOVSSrr (v4i32 (V_SET0PI)),
3739 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3742 // Splat v2f64 / v2i64
3743 let AddedComplexity = 10 in {
3744 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3745 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3746 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3747 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3748 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3749 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3750 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3751 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3754 // Special unary SHUFPSrri case.
3755 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3756 (SHUFPSrri VR128:$src1, VR128:$src1,
3757 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3758 let AddedComplexity = 5 in
3759 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3760 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3761 Requires<[HasSSE2]>;
3762 // Special unary SHUFPDrri case.
3763 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3764 (SHUFPDrri VR128:$src1, VR128:$src1,
3765 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3766 Requires<[HasSSE2]>;
3767 // Special unary SHUFPDrri case.
3768 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3769 (SHUFPDrri VR128:$src1, VR128:$src1,
3770 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3771 Requires<[HasSSE2]>;
3772 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3773 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3774 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3775 Requires<[HasSSE2]>;
3777 // Special binary v4i32 shuffle cases with SHUFPS.
3778 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3779 (SHUFPSrri VR128:$src1, VR128:$src2,
3780 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3781 Requires<[HasSSE2]>;
3782 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3783 (SHUFPSrmi VR128:$src1, addr:$src2,
3784 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3785 Requires<[HasSSE2]>;
3786 // Special binary v2i64 shuffle cases using SHUFPDrri.
3787 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3788 (SHUFPDrri VR128:$src1, VR128:$src2,
3789 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3790 Requires<[HasSSE2]>;
3792 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3793 let AddedComplexity = 15 in {
3794 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3795 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3796 Requires<[OptForSpeed, HasSSE2]>;
3797 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3798 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3799 Requires<[OptForSpeed, HasSSE2]>;
3801 let AddedComplexity = 10 in {
3802 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3803 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3804 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3805 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3806 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3807 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3808 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3809 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3812 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3813 let AddedComplexity = 15 in {
3814 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3815 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3816 Requires<[OptForSpeed, HasSSE2]>;
3817 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3818 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3819 Requires<[OptForSpeed, HasSSE2]>;
3821 let AddedComplexity = 10 in {
3822 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3823 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3824 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3825 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3826 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3827 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3828 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3829 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3832 let AddedComplexity = 20 in {
3833 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3834 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3835 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3837 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3838 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3839 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3841 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3842 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3843 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3844 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3845 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3848 let AddedComplexity = 20 in {
3849 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3850 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3851 (MOVLPSrm VR128:$src1, addr:$src2)>;
3852 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3853 (MOVLPDrm VR128:$src1, addr:$src2)>;
3854 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3855 (MOVLPSrm VR128:$src1, addr:$src2)>;
3856 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3857 (MOVLPDrm VR128:$src1, addr:$src2)>;
3860 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3861 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3862 (MOVLPSmr addr:$src1, VR128:$src2)>;
3863 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3864 (MOVLPDmr addr:$src1, VR128:$src2)>;
3865 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3867 (MOVLPSmr addr:$src1, VR128:$src2)>;
3868 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3869 (MOVLPDmr addr:$src1, VR128:$src2)>;
3871 let AddedComplexity = 15 in {
3872 // Setting the lowest element in the vector.
3873 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3874 (MOVSSrr (v4i32 VR128:$src1),
3875 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3876 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3877 (MOVSDrr (v2i64 VR128:$src1),
3878 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3880 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3881 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3882 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3883 Requires<[HasSSE2]>;
3884 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3885 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3886 Requires<[HasSSE2]>;
3889 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3890 // fall back to this for SSE1)
3891 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3892 (SHUFPSrri VR128:$src2, VR128:$src1,
3893 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3895 // Set lowest element and zero upper elements.
3896 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3897 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3899 // Some special case pandn patterns.
3900 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3902 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3903 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3905 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3906 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3908 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3910 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3911 (memop addr:$src2))),
3912 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3913 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3914 (memop addr:$src2))),
3915 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3916 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3917 (memop addr:$src2))),
3918 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3920 // vector -> vector casts
3921 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3922 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3923 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3924 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3926 // Use movaps / movups for SSE integer load / store (one byte shorter).
3927 let Predicates = [HasSSE1] in {
3928 def : Pat<(alignedloadv4i32 addr:$src),
3929 (MOVAPSrm addr:$src)>;
3930 def : Pat<(loadv4i32 addr:$src),
3931 (MOVUPSrm addr:$src)>;
3932 def : Pat<(alignedloadv2i64 addr:$src),
3933 (MOVAPSrm addr:$src)>;
3934 def : Pat<(loadv2i64 addr:$src),
3935 (MOVUPSrm addr:$src)>;
3937 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3938 (MOVAPSmr addr:$dst, VR128:$src)>;
3939 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3940 (MOVAPSmr addr:$dst, VR128:$src)>;
3941 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3942 (MOVAPSmr addr:$dst, VR128:$src)>;
3943 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3944 (MOVAPSmr addr:$dst, VR128:$src)>;
3945 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3946 (MOVUPSmr addr:$dst, VR128:$src)>;
3947 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3948 (MOVUPSmr addr:$dst, VR128:$src)>;
3949 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3950 (MOVUPSmr addr:$dst, VR128:$src)>;
3951 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3952 (MOVUPSmr addr:$dst, VR128:$src)>;
3955 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3956 let Predicates = [HasAVX] in {
3957 def : Pat<(alignedloadv4i32 addr:$src),
3958 (VMOVAPSrm addr:$src)>;
3959 def : Pat<(loadv4i32 addr:$src),
3960 (VMOVUPSrm addr:$src)>;
3961 def : Pat<(alignedloadv2i64 addr:$src),
3962 (VMOVAPSrm addr:$src)>;
3963 def : Pat<(loadv2i64 addr:$src),
3964 (VMOVUPSrm addr:$src)>;
3966 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3967 (VMOVAPSmr addr:$dst, VR128:$src)>;
3968 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3969 (VMOVAPSmr addr:$dst, VR128:$src)>;
3970 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3971 (VMOVAPSmr addr:$dst, VR128:$src)>;
3972 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3973 (VMOVAPSmr addr:$dst, VR128:$src)>;
3974 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3975 (VMOVUPSmr addr:$dst, VR128:$src)>;
3976 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3977 (VMOVUPSmr addr:$dst, VR128:$src)>;
3978 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3979 (VMOVUPSmr addr:$dst, VR128:$src)>;
3980 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3981 (VMOVUPSmr addr:$dst, VR128:$src)>;
3984 //===----------------------------------------------------------------------===//
3985 // SSE4.1 - Packed Move with Sign/Zero Extend
3986 //===----------------------------------------------------------------------===//
3988 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3989 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3990 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3991 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3993 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3994 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3996 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4000 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4001 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4003 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4005 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4007 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4009 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4011 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4015 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4016 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4017 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4018 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4019 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4020 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4022 // Common patterns involving scalar load.
4023 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4024 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4025 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4026 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4028 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4029 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4030 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4031 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4033 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4034 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4035 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4036 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4038 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4039 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4040 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4041 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4043 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4044 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4045 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4046 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4048 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4049 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4050 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4051 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4054 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4055 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4056 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4057 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4059 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4060 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4062 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4066 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4067 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4069 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4071 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4073 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4077 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4078 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4079 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4080 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4082 // Common patterns involving scalar load
4083 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4084 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4085 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4086 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4088 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4089 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4090 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4091 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4094 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4095 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4096 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4097 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4099 // Expecting a i16 load any extended to i32 value.
4100 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4101 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4102 [(set VR128:$dst, (IntId (bitconvert
4103 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4107 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4108 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4110 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4113 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4114 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4116 // Common patterns involving scalar load
4117 def : Pat<(int_x86_sse41_pmovsxbq
4118 (bitconvert (v4i32 (X86vzmovl
4119 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4120 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4122 def : Pat<(int_x86_sse41_pmovzxbq
4123 (bitconvert (v4i32 (X86vzmovl
4124 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4125 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4127 //===----------------------------------------------------------------------===//
4128 // SSE4.1 - Extract Instructions
4129 //===----------------------------------------------------------------------===//
4131 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4132 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4133 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4134 (ins VR128:$src1, i32i8imm:$src2),
4135 !strconcat(OpcodeStr,
4136 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4137 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4139 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4140 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4141 !strconcat(OpcodeStr,
4142 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4145 // There's an AssertZext in the way of writing the store pattern
4146 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4149 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4150 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4151 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4152 (ins VR128:$src1, i32i8imm:$src2),
4153 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4156 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4159 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4160 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4161 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4162 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4163 !strconcat(OpcodeStr,
4164 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4167 // There's an AssertZext in the way of writing the store pattern
4168 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4171 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4172 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4174 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4177 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4178 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4179 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4180 (ins VR128:$src1, i32i8imm:$src2),
4181 !strconcat(OpcodeStr,
4182 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4184 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4185 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4186 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4187 !strconcat(OpcodeStr,
4188 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4189 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4190 addr:$dst)]>, OpSize;
4193 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4194 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4196 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4198 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4199 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4200 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4201 (ins VR128:$src1, i32i8imm:$src2),
4202 !strconcat(OpcodeStr,
4203 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4205 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4206 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4207 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4208 !strconcat(OpcodeStr,
4209 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4210 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4211 addr:$dst)]>, OpSize, REX_W;
4214 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4215 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4217 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4219 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4221 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4222 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4223 (ins VR128:$src1, i32i8imm:$src2),
4224 !strconcat(OpcodeStr,
4225 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4227 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4229 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4230 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4231 !strconcat(OpcodeStr,
4232 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4233 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4234 addr:$dst)]>, OpSize;
4237 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4238 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4239 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4240 (ins VR128:$src1, i32i8imm:$src2),
4241 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4244 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4246 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4247 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4250 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4251 Requires<[HasSSE41]>;
4253 //===----------------------------------------------------------------------===//
4254 // SSE4.1 - Insert Instructions
4255 //===----------------------------------------------------------------------===//
4257 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4258 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4259 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4261 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4263 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4265 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4266 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4267 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4269 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4271 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4273 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4274 imm:$src3))]>, OpSize;
4277 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4278 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4279 let Constraints = "$src1 = $dst" in
4280 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4282 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4283 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4284 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4286 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4288 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4290 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4292 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4293 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4295 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4297 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4299 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4300 imm:$src3)))]>, OpSize;
4303 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4304 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4305 let Constraints = "$src1 = $dst" in
4306 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4308 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4309 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4310 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4312 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4314 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4316 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4318 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4319 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4321 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4323 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4325 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4326 imm:$src3)))]>, OpSize;
4329 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4330 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4331 let Constraints = "$src1 = $dst" in
4332 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4334 // insertps has a few different modes, there's the first two here below which
4335 // are optimized inserts that won't zero arbitrary elements in the destination
4336 // vector. The next one matches the intrinsic and could zero arbitrary elements
4337 // in the target vector.
4338 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4339 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4340 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4342 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4344 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4346 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4348 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4349 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4351 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4353 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4355 (X86insrtps VR128:$src1,
4356 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4357 imm:$src3))]>, OpSize;
4360 let Constraints = "$src1 = $dst" in
4361 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4362 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4363 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4365 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4366 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4368 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4369 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4370 Requires<[HasSSE41]>;
4372 //===----------------------------------------------------------------------===//
4373 // SSE4.1 - Round Instructions
4374 //===----------------------------------------------------------------------===//
4376 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4377 X86MemOperand x86memop, RegisterClass RC,
4378 PatFrag mem_frag32, PatFrag mem_frag64,
4379 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4380 // Intrinsic operation, reg.
4381 // Vector intrinsic operation, reg
4382 def PSr : SS4AIi8<opcps, MRMSrcReg,
4383 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4384 !strconcat(OpcodeStr,
4385 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4386 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4389 // Vector intrinsic operation, mem
4390 def PSm : Ii8<opcps, MRMSrcMem,
4391 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4392 !strconcat(OpcodeStr,
4393 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4395 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4397 Requires<[HasSSE41]>;
4399 // Vector intrinsic operation, reg
4400 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4401 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4402 !strconcat(OpcodeStr,
4403 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4404 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4407 // Vector intrinsic operation, mem
4408 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4409 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4410 !strconcat(OpcodeStr,
4411 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4413 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4417 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4418 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4419 // Intrinsic operation, reg.
4420 // Vector intrinsic operation, reg
4421 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4422 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4423 !strconcat(OpcodeStr,
4424 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4427 // Vector intrinsic operation, mem
4428 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4429 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4430 !strconcat(OpcodeStr,
4431 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4432 []>, TA, OpSize, Requires<[HasSSE41]>;
4434 // Vector intrinsic operation, reg
4435 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4436 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4437 !strconcat(OpcodeStr,
4438 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4441 // Vector intrinsic operation, mem
4442 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4443 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4444 !strconcat(OpcodeStr,
4445 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4449 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4452 Intrinsic F64Int, bit Is2Addr = 1> {
4453 // Intrinsic operation, reg.
4454 def SSr : SS4AIi8<opcss, MRMSrcReg,
4455 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4457 !strconcat(OpcodeStr,
4458 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4459 !strconcat(OpcodeStr,
4460 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4461 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4464 // Intrinsic operation, mem.
4465 def SSm : SS4AIi8<opcss, MRMSrcMem,
4466 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4468 !strconcat(OpcodeStr,
4469 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4470 !strconcat(OpcodeStr,
4471 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4473 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4476 // Intrinsic operation, reg.
4477 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4478 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4480 !strconcat(OpcodeStr,
4481 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4482 !strconcat(OpcodeStr,
4483 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4484 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4487 // Intrinsic operation, mem.
4488 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4489 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4491 !strconcat(OpcodeStr,
4492 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4493 !strconcat(OpcodeStr,
4494 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4496 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4500 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4502 // Intrinsic operation, reg.
4503 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4504 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4505 !strconcat(OpcodeStr,
4506 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4509 // Intrinsic operation, mem.
4510 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4511 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4512 !strconcat(OpcodeStr,
4513 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4516 // Intrinsic operation, reg.
4517 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4518 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4519 !strconcat(OpcodeStr,
4520 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4523 // Intrinsic operation, mem.
4524 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4525 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4526 !strconcat(OpcodeStr,
4527 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4531 // FP round - roundss, roundps, roundsd, roundpd
4532 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4534 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4535 memopv4f32, memopv2f64,
4536 int_x86_sse41_round_ps,
4537 int_x86_sse41_round_pd>, VEX;
4538 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4539 memopv8f32, memopv4f64,
4540 int_x86_avx_round_ps_256,
4541 int_x86_avx_round_pd_256>, VEX;
4542 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4543 int_x86_sse41_round_ss,
4544 int_x86_sse41_round_sd, 0>, VEX_4V;
4546 // Instructions for the assembler
4547 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4549 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4551 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4554 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4555 memopv4f32, memopv2f64,
4556 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4557 let Constraints = "$src1 = $dst" in
4558 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4559 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4561 //===----------------------------------------------------------------------===//
4562 // SSE4.1 - Packed Bit Test
4563 //===----------------------------------------------------------------------===//
4565 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4566 // the intel intrinsic that corresponds to this.
4567 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4568 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4569 "vptest\t{$src2, $src1|$src1, $src2}",
4570 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4572 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4573 "vptest\t{$src2, $src1|$src1, $src2}",
4574 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4577 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4578 "vptest\t{$src2, $src1|$src1, $src2}",
4579 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4581 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4582 "vptest\t{$src2, $src1|$src1, $src2}",
4583 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4587 let Defs = [EFLAGS] in {
4588 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4589 "ptest \t{$src2, $src1|$src1, $src2}",
4590 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4592 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4593 "ptest \t{$src2, $src1|$src1, $src2}",
4594 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4598 // The bit test instructions below are AVX only
4599 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4600 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4601 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4602 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4603 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4604 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4605 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4606 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4610 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4611 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4612 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4613 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4614 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4617 //===----------------------------------------------------------------------===//
4618 // SSE4.1 - Misc Instructions
4619 //===----------------------------------------------------------------------===//
4621 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4622 "popcnt{w}\t{$src, $dst|$dst, $src}",
4623 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4624 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4625 "popcnt{w}\t{$src, $dst|$dst, $src}",
4626 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4628 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4629 "popcnt{l}\t{$src, $dst|$dst, $src}",
4630 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4631 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4632 "popcnt{l}\t{$src, $dst|$dst, $src}",
4633 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4635 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4636 "popcnt{q}\t{$src, $dst|$dst, $src}",
4637 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4638 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4639 "popcnt{q}\t{$src, $dst|$dst, $src}",
4640 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4644 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4645 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4646 Intrinsic IntId128> {
4647 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4649 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4650 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4651 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4653 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4656 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4659 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4660 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4661 int_x86_sse41_phminposuw>, VEX;
4662 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4663 int_x86_sse41_phminposuw>;
4665 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4666 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4667 Intrinsic IntId128, bit Is2Addr = 1> {
4668 let isCommutable = 1 in
4669 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4670 (ins VR128:$src1, VR128:$src2),
4672 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4673 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4674 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4675 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4676 (ins VR128:$src1, i128mem:$src2),
4678 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4679 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4681 (IntId128 VR128:$src1,
4682 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4685 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4686 let isCommutable = 0 in
4687 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4689 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4691 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4693 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4695 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4697 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4699 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4701 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4703 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4705 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4707 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4711 let Constraints = "$src1 = $dst" in {
4712 let isCommutable = 0 in
4713 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4714 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4715 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4716 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4717 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4718 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4719 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4720 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4721 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4722 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4723 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4726 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4727 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4728 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4729 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4731 /// SS48I_binop_rm - Simple SSE41 binary operator.
4732 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4733 ValueType OpVT, bit Is2Addr = 1> {
4734 let isCommutable = 1 in
4735 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4736 (ins VR128:$src1, VR128:$src2),
4738 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4739 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4740 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4742 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4743 (ins VR128:$src1, i128mem:$src2),
4745 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4746 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4747 [(set VR128:$dst, (OpNode VR128:$src1,
4748 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4752 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4753 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4754 let Constraints = "$src1 = $dst" in
4755 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4757 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4758 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4759 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4760 X86MemOperand x86memop, bit Is2Addr = 1> {
4761 let isCommutable = 1 in
4762 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4763 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4765 !strconcat(OpcodeStr,
4766 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4767 !strconcat(OpcodeStr,
4768 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4769 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4771 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4772 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4774 !strconcat(OpcodeStr,
4775 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4776 !strconcat(OpcodeStr,
4777 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4780 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4784 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4785 let isCommutable = 0 in {
4786 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4787 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4788 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4789 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4790 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4791 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4792 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4793 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4794 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4795 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4796 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4797 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4799 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4800 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4801 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4802 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4803 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4804 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4807 let Constraints = "$src1 = $dst" in {
4808 let isCommutable = 0 in {
4809 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4810 VR128, memopv16i8, i128mem>;
4811 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4812 VR128, memopv16i8, i128mem>;
4813 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4814 VR128, memopv16i8, i128mem>;
4815 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4816 VR128, memopv16i8, i128mem>;
4818 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4819 VR128, memopv16i8, i128mem>;
4820 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4821 VR128, memopv16i8, i128mem>;
4824 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4825 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4826 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4827 RegisterClass RC, X86MemOperand x86memop,
4828 PatFrag mem_frag, Intrinsic IntId> {
4829 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4830 (ins RC:$src1, RC:$src2, RC:$src3),
4831 !strconcat(OpcodeStr,
4832 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4833 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4834 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4836 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4837 (ins RC:$src1, x86memop:$src2, RC:$src3),
4838 !strconcat(OpcodeStr,
4839 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4841 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4843 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4847 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4848 memopv16i8, int_x86_sse41_blendvpd>;
4849 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4850 memopv16i8, int_x86_sse41_blendvps>;
4851 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4852 memopv16i8, int_x86_sse41_pblendvb>;
4853 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4854 memopv32i8, int_x86_avx_blendv_pd_256>;
4855 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4856 memopv32i8, int_x86_avx_blendv_ps_256>;
4858 /// SS41I_ternary_int - SSE 4.1 ternary operator
4859 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4860 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4861 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4862 (ins VR128:$src1, VR128:$src2),
4863 !strconcat(OpcodeStr,
4864 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4865 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4868 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4869 (ins VR128:$src1, i128mem:$src2),
4870 !strconcat(OpcodeStr,
4871 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4874 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4878 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4879 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4880 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4882 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4883 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4884 "vmovntdqa\t{$src, $dst|$dst, $src}",
4885 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4887 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4888 "movntdqa\t{$src, $dst|$dst, $src}",
4889 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4892 //===----------------------------------------------------------------------===//
4893 // SSE4.2 - Compare Instructions
4894 //===----------------------------------------------------------------------===//
4896 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4897 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4898 Intrinsic IntId128, bit Is2Addr = 1> {
4899 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4900 (ins VR128:$src1, VR128:$src2),
4902 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4903 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4904 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4906 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4907 (ins VR128:$src1, i128mem:$src2),
4909 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4910 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4912 (IntId128 VR128:$src1,
4913 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4916 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4917 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4919 let Constraints = "$src1 = $dst" in
4920 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4922 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4923 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4924 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4925 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4927 //===----------------------------------------------------------------------===//
4928 // SSE4.2 - String/text Processing Instructions
4929 //===----------------------------------------------------------------------===//
4931 // Packed Compare Implicit Length Strings, Return Mask
4932 multiclass pseudo_pcmpistrm<string asm> {
4933 def REG : PseudoI<(outs VR128:$dst),
4934 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4935 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4937 def MEM : PseudoI<(outs VR128:$dst),
4938 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4939 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4940 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4943 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4944 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4945 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4948 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4949 Predicates = [HasAVX] in {
4950 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4951 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4952 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4953 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4954 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4955 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4958 let Defs = [XMM0, EFLAGS] in {
4959 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4960 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4961 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4962 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4963 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4964 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4967 // Packed Compare Explicit Length Strings, Return Mask
4968 multiclass pseudo_pcmpestrm<string asm> {
4969 def REG : PseudoI<(outs VR128:$dst),
4970 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4971 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4972 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4973 def MEM : PseudoI<(outs VR128:$dst),
4974 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4975 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4976 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4979 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4980 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4981 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4984 let isAsmParserOnly = 1, Predicates = [HasAVX],
4985 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4986 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4987 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4988 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4989 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4990 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4991 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4994 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4995 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4996 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4997 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4998 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4999 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5000 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5003 // Packed Compare Implicit Length Strings, Return Index
5004 let Defs = [ECX, EFLAGS] in {
5005 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5006 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5007 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5008 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5009 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5010 (implicit EFLAGS)]>, OpSize;
5011 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5012 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5013 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5014 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5015 (implicit EFLAGS)]>, OpSize;
5019 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5020 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5022 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5024 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5026 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5028 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5030 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5034 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5035 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5036 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5037 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5038 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5039 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5041 // Packed Compare Explicit Length Strings, Return Index
5042 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5043 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5044 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5045 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5046 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5047 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5048 (implicit EFLAGS)]>, OpSize;
5049 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5050 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5051 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5053 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5054 (implicit EFLAGS)]>, OpSize;
5058 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5059 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5061 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5063 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5065 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5067 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5069 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5073 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5074 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5075 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5076 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5077 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5078 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5080 //===----------------------------------------------------------------------===//
5081 // SSE4.2 - CRC Instructions
5082 //===----------------------------------------------------------------------===//
5084 // No CRC instructions have AVX equivalents
5086 // crc intrinsic instruction
5087 // This set of instructions are only rm, the only difference is the size
5089 let Constraints = "$src1 = $dst" in {
5090 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5091 (ins GR32:$src1, i8mem:$src2),
5092 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5094 (int_x86_sse42_crc32_8 GR32:$src1,
5095 (load addr:$src2)))]>;
5096 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5097 (ins GR32:$src1, GR8:$src2),
5098 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5100 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5101 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5102 (ins GR32:$src1, i16mem:$src2),
5103 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5105 (int_x86_sse42_crc32_16 GR32:$src1,
5106 (load addr:$src2)))]>,
5108 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5109 (ins GR32:$src1, GR16:$src2),
5110 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5112 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5114 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5115 (ins GR32:$src1, i32mem:$src2),
5116 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5118 (int_x86_sse42_crc32_32 GR32:$src1,
5119 (load addr:$src2)))]>;
5120 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5121 (ins GR32:$src1, GR32:$src2),
5122 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5124 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5125 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5126 (ins GR64:$src1, i8mem:$src2),
5127 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5129 (int_x86_sse42_crc64_8 GR64:$src1,
5130 (load addr:$src2)))]>,
5132 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5133 (ins GR64:$src1, GR8:$src2),
5134 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5136 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5138 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5139 (ins GR64:$src1, i64mem:$src2),
5140 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5142 (int_x86_sse42_crc64_64 GR64:$src1,
5143 (load addr:$src2)))]>,
5145 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5146 (ins GR64:$src1, GR64:$src2),
5147 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5149 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5153 //===----------------------------------------------------------------------===//
5154 // AES-NI Instructions
5155 //===----------------------------------------------------------------------===//
5157 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5158 Intrinsic IntId128, bit Is2Addr = 1> {
5159 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5160 (ins VR128:$src1, VR128:$src2),
5162 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5163 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5164 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5166 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5167 (ins VR128:$src1, i128mem:$src2),
5169 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5170 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5172 (IntId128 VR128:$src1,
5173 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5176 // Perform One Round of an AES Encryption/Decryption Flow
5177 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5178 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5179 int_x86_aesni_aesenc, 0>, VEX_4V;
5180 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5181 int_x86_aesni_aesenclast, 0>, VEX_4V;
5182 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5183 int_x86_aesni_aesdec, 0>, VEX_4V;
5184 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5185 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5188 let Constraints = "$src1 = $dst" in {
5189 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5190 int_x86_aesni_aesenc>;
5191 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5192 int_x86_aesni_aesenclast>;
5193 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5194 int_x86_aesni_aesdec>;
5195 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5196 int_x86_aesni_aesdeclast>;
5199 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5200 (AESENCrr VR128:$src1, VR128:$src2)>;
5201 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5202 (AESENCrm VR128:$src1, addr:$src2)>;
5203 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5204 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5205 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5206 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5207 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5208 (AESDECrr VR128:$src1, VR128:$src2)>;
5209 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5210 (AESDECrm VR128:$src1, addr:$src2)>;
5211 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5212 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5213 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5214 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5216 // Perform the AES InvMixColumn Transformation
5217 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5218 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5220 "vaesimc\t{$src1, $dst|$dst, $src1}",
5222 (int_x86_aesni_aesimc VR128:$src1))]>,
5224 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5225 (ins i128mem:$src1),
5226 "vaesimc\t{$src1, $dst|$dst, $src1}",
5228 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5231 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5233 "aesimc\t{$src1, $dst|$dst, $src1}",
5235 (int_x86_aesni_aesimc VR128:$src1))]>,
5237 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5238 (ins i128mem:$src1),
5239 "aesimc\t{$src1, $dst|$dst, $src1}",
5241 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5244 // AES Round Key Generation Assist
5245 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5246 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5247 (ins VR128:$src1, i8imm:$src2),
5248 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5250 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5252 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5253 (ins i128mem:$src1, i8imm:$src2),
5254 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5256 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5260 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5261 (ins VR128:$src1, i8imm:$src2),
5262 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5264 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5266 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5267 (ins i128mem:$src1, i8imm:$src2),
5268 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5270 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5274 //===----------------------------------------------------------------------===//
5275 // CLMUL Instructions
5276 //===----------------------------------------------------------------------===//
5278 // Only the AVX version of CLMUL instructions are described here.
5280 // Carry-less Multiplication instructions
5281 let isAsmParserOnly = 1 in {
5282 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5283 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5284 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5287 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5288 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5289 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5293 multiclass avx_vpclmul<string asm> {
5294 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5295 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5298 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5299 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5302 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5303 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5304 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5305 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5307 } // isAsmParserOnly
5309 //===----------------------------------------------------------------------===//
5311 //===----------------------------------------------------------------------===//
5313 let isAsmParserOnly = 1 in {
5315 // Load from memory and broadcast to all elements of the destination operand
5316 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5317 X86MemOperand x86memop, Intrinsic Int> :
5318 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5319 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5320 [(set RC:$dst, (Int addr:$src))]>, VEX;
5322 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5323 int_x86_avx_vbroadcastss>;
5324 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5325 int_x86_avx_vbroadcastss_256>;
5326 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5327 int_x86_avx_vbroadcast_sd_256>;
5328 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5329 int_x86_avx_vbroadcastf128_pd_256>;
5331 // Insert packed floating-point values
5332 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5333 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5334 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5336 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5337 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5338 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5341 // Extract packed floating-point values
5342 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5343 (ins VR256:$src1, i8imm:$src2),
5344 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5346 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5347 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5348 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5351 // Conditional SIMD Packed Loads and Stores
5352 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5353 Intrinsic IntLd, Intrinsic IntLd256,
5354 Intrinsic IntSt, Intrinsic IntSt256,
5355 PatFrag pf128, PatFrag pf256> {
5356 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5357 (ins VR128:$src1, f128mem:$src2),
5358 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5359 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5361 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5362 (ins VR256:$src1, f256mem:$src2),
5363 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5364 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5366 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5367 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5368 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5369 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5370 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5371 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5372 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5373 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5376 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5377 int_x86_avx_maskload_ps,
5378 int_x86_avx_maskload_ps_256,
5379 int_x86_avx_maskstore_ps,
5380 int_x86_avx_maskstore_ps_256,
5381 memopv4f32, memopv8f32>;
5382 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5383 int_x86_avx_maskload_pd,
5384 int_x86_avx_maskload_pd_256,
5385 int_x86_avx_maskstore_pd,
5386 int_x86_avx_maskstore_pd_256,
5387 memopv2f64, memopv4f64>;
5389 // Permute Floating-Point Values
5390 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5391 RegisterClass RC, X86MemOperand x86memop_f,
5392 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5393 Intrinsic IntVar, Intrinsic IntImm> {
5394 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5395 (ins RC:$src1, RC:$src2),
5396 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5397 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5398 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5399 (ins RC:$src1, x86memop_i:$src2),
5400 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5401 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5403 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5404 (ins RC:$src1, i8imm:$src2),
5405 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5406 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5407 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5408 (ins x86memop_f:$src1, i8imm:$src2),
5409 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5410 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5413 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5414 memopv4f32, memopv4i32,
5415 int_x86_avx_vpermilvar_ps,
5416 int_x86_avx_vpermil_ps>;
5417 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5418 memopv8f32, memopv8i32,
5419 int_x86_avx_vpermilvar_ps_256,
5420 int_x86_avx_vpermil_ps_256>;
5421 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5422 memopv2f64, memopv2i64,
5423 int_x86_avx_vpermilvar_pd,
5424 int_x86_avx_vpermil_pd>;
5425 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5426 memopv4f64, memopv4i64,
5427 int_x86_avx_vpermilvar_pd_256,
5428 int_x86_avx_vpermil_pd_256>;
5430 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5431 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5432 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5434 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5435 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5436 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5439 // Zero All YMM registers
5440 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5441 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5443 // Zero Upper bits of YMM registers
5444 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5445 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5447 } // isAsmParserOnly
5449 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5450 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5451 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5452 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5453 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5454 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5456 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5457 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5458 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5459 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5460 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5461 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5463 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5464 (VBROADCASTF128 addr:$src)>;
5466 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5467 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5468 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5469 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5470 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5471 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5473 def : Pat<(int_x86_avx_vperm2f128_ps_256
5474 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5475 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5476 def : Pat<(int_x86_avx_vperm2f128_pd_256
5477 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5478 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5479 def : Pat<(int_x86_avx_vperm2f128_si_256
5480 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5481 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5483 //===----------------------------------------------------------------------===//
5484 // SSE Shuffle pattern fragments
5485 //===----------------------------------------------------------------------===//
5487 // This is part of a "work in progress" refactoring. The idea is that all
5488 // vector shuffles are going to be translated into target specific nodes and
5489 // directly matched by the patterns below (which can be changed along the way)
5490 // The AVX version of some but not all of them are described here, and more
5491 // should come in a near future.
5493 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5494 // SSE2 loads, which are always promoted to v2i64. The last one should match
5495 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5496 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5497 // we investigate further.
5498 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5500 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5501 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5503 (PSHUFDmi addr:$src1, imm:$imm)>;
5504 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5506 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5508 // Shuffle with PSHUFD instruction.
5509 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5510 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5511 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5512 (PSHUFDri VR128:$src1, imm:$imm)>;
5514 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5515 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5516 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5517 (PSHUFDri VR128:$src1, imm:$imm)>;
5519 // Shuffle with SHUFPD instruction.
5520 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5521 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5522 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5523 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5524 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5525 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5527 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5528 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5529 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5530 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5532 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5533 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5534 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5535 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5537 // Shuffle with SHUFPS instruction.
5538 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5539 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5540 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5541 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5542 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5543 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5545 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5546 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5547 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5548 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5550 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5551 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5552 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5553 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5554 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5555 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5557 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5558 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5559 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5560 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5562 // Shuffle with MOVHLPS instruction
5563 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5564 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5565 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5566 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5568 // Shuffle with MOVDDUP instruction
5569 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5570 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5571 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5572 (MOVDDUPrm addr:$src)>;
5574 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5575 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5576 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5577 (MOVDDUPrm addr:$src)>;
5579 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5580 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5581 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5582 (MOVDDUPrm addr:$src)>;
5584 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5585 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5586 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5587 (MOVDDUPrm addr:$src)>;
5589 def : Pat<(X86Movddup (bc_v2f64
5590 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5591 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5592 def : Pat<(X86Movddup (bc_v2f64
5593 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5594 (MOVDDUPrm addr:$src)>;
5597 // Shuffle with UNPCKLPS
5598 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5599 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5600 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5601 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5603 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5604 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5605 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5606 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5608 // Shuffle with UNPCKHPS
5609 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5610 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5611 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5612 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5614 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5615 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5616 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5617 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5619 // Shuffle with UNPCKLPD
5620 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5621 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5622 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5623 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5625 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5626 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5627 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5628 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5630 // Shuffle with UNPCKHPD
5631 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5632 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5633 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5634 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5636 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5637 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5638 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5639 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5641 // Shuffle with PUNPCKLBW
5642 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5643 (bc_v16i8 (memopv2i64 addr:$src2)))),
5644 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5645 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5646 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5648 // Shuffle with PUNPCKLWD
5649 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5650 (bc_v8i16 (memopv2i64 addr:$src2)))),
5651 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5652 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5653 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5655 // Shuffle with PUNPCKLDQ
5656 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5657 (bc_v4i32 (memopv2i64 addr:$src2)))),
5658 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5659 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5660 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5662 // Shuffle with PUNPCKLQDQ
5663 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5664 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5665 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5666 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5668 // Shuffle with PUNPCKHBW
5669 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5670 (bc_v16i8 (memopv2i64 addr:$src2)))),
5671 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5672 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5673 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5675 // Shuffle with PUNPCKHWD
5676 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5677 (bc_v8i16 (memopv2i64 addr:$src2)))),
5678 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5679 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5680 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5682 // Shuffle with PUNPCKHDQ
5683 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5684 (bc_v4i32 (memopv2i64 addr:$src2)))),
5685 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5686 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5687 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5689 // Shuffle with PUNPCKHQDQ
5690 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5691 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5692 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5693 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5695 // Shuffle with MOVLHPS
5696 def : Pat<(X86Movlhps VR128:$src1,
5697 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5698 (MOVHPSrm VR128:$src1, addr:$src2)>;
5699 def : Pat<(X86Movlhps VR128:$src1,
5700 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5701 (MOVHPSrm VR128:$src1, addr:$src2)>;
5702 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5703 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5704 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5705 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5706 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5707 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5709 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5710 // is during lowering, where it's not possible to recognize the load fold cause
5711 // it has two uses through a bitcast. One use disappears at isel time and the
5712 // fold opportunity reappears.
5713 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5714 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5716 // Shuffle with MOVLHPD
5717 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5718 (scalar_to_vector (loadf64 addr:$src2)))),
5719 (MOVHPDrm VR128:$src1, addr:$src2)>;
5721 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5722 // is during lowering, where it's not possible to recognize the load fold cause
5723 // it has two uses through a bitcast. One use disappears at isel time and the
5724 // fold opportunity reappears.
5725 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5726 (scalar_to_vector (loadf64 addr:$src2)))),
5727 (MOVHPDrm VR128:$src1, addr:$src2)>;
5729 // Shuffle with MOVSS
5730 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5731 (MOVSSrr VR128:$src1, FR32:$src2)>;
5732 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5733 (MOVSSrr (v4i32 VR128:$src1),
5734 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5735 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5736 (MOVSSrr (v4f32 VR128:$src1),
5737 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5738 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5739 // is during lowering, where it's not possible to recognize the load fold cause
5740 // it has two uses through a bitcast. One use disappears at isel time and the
5741 // fold opportunity reappears.
5742 def : Pat<(X86Movss VR128:$src1,
5743 (bc_v4i32 (v2i64 (load addr:$src2)))),
5744 (MOVLPSrm VR128:$src1, addr:$src2)>;
5746 // Shuffle with MOVSD
5747 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5748 (MOVSDrr VR128:$src1, FR64:$src2)>;
5749 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5750 (MOVSDrr (v2i64 VR128:$src1),
5751 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5752 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5753 (MOVSDrr (v2f64 VR128:$src1),
5754 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5755 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5756 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5757 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5758 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5760 // Shuffle with MOVSHDUP
5761 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5762 (MOVSHDUPrr VR128:$src)>;
5763 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5764 (MOVSHDUPrm addr:$src)>;
5766 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5767 (MOVSHDUPrr VR128:$src)>;
5768 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5769 (MOVSHDUPrm addr:$src)>;
5771 // Shuffle with MOVSLDUP
5772 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5773 (MOVSLDUPrr VR128:$src)>;
5774 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5775 (MOVSLDUPrm addr:$src)>;
5777 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5778 (MOVSLDUPrr VR128:$src)>;
5779 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5780 (MOVSLDUPrm addr:$src)>;
5782 // Shuffle with PSHUFHW
5783 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5784 (PSHUFHWri VR128:$src, imm:$imm)>;
5785 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5786 (PSHUFHWmi addr:$src, imm:$imm)>;
5788 // Shuffle with PSHUFLW
5789 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5790 (PSHUFLWri VR128:$src, imm:$imm)>;
5791 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5792 (PSHUFLWmi addr:$src, imm:$imm)>;
5794 // Shuffle with PALIGN
5795 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5796 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5797 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5798 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5799 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5800 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5801 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5802 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5804 // Shuffle with MOVLPS
5805 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5806 (MOVLPSrm VR128:$src1, addr:$src2)>;
5807 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5808 (MOVLPSrm VR128:$src1, addr:$src2)>;
5809 def : Pat<(X86Movlps VR128:$src1,
5810 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5811 (MOVLPSrm VR128:$src1, addr:$src2)>;
5812 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5813 // is during lowering, where it's not possible to recognize the load fold cause
5814 // it has two uses through a bitcast. One use disappears at isel time and the
5815 // fold opportunity reappears.
5816 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5817 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5819 // Shuffle with MOVLPD
5820 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5821 (MOVLPDrm VR128:$src1, addr:$src2)>;
5822 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5823 (MOVLPDrm VR128:$src1, addr:$src2)>;
5824 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5825 (scalar_to_vector (loadf64 addr:$src2)))),
5826 (MOVLPDrm VR128:$src1, addr:$src2)>;
5828 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5829 def : Pat<(store (f64 (vector_extract
5830 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5831 (MOVHPSmr addr:$dst, VR128:$src)>;
5832 def : Pat<(store (f64 (vector_extract
5833 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5834 (MOVHPDmr addr:$dst, VR128:$src)>;
5836 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5837 (MOVLPSmr addr:$src1, VR128:$src2)>;
5838 def : Pat<(store (v4i32 (X86Movlps
5839 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5840 (MOVLPSmr addr:$src1, VR128:$src2)>;
5842 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5843 (MOVLPDmr addr:$src1, VR128:$src2)>;
5844 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5845 (MOVLPDmr addr:$src1, VR128:$src2)>;