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)>;
717 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
718 "cvtsd2ss\t{$src, $dst|$dst, $src}",
719 [(set FR32:$dst, (fround FR64:$src))]>;
720 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
721 "cvtsd2ss\t{$src, $dst|$dst, $src}",
722 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
723 Requires<[HasSSE2, OptForSize]>;
725 let isAsmParserOnly = 1 in
726 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
727 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
729 let Constraints = "$src1 = $dst" in
730 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
731 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
733 // Convert scalar single to scalar double
734 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
735 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
736 (ins FR32:$src1, FR32:$src2),
737 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
738 []>, XS, Requires<[HasAVX]>, VEX_4V;
739 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
740 (ins FR32:$src1, f32mem:$src2),
741 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
742 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
744 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>;
746 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
747 "cvtss2sd\t{$src, $dst|$dst, $src}",
748 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
750 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
751 "cvtss2sd\t{$src, $dst|$dst, $src}",
752 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
753 Requires<[HasSSE2, OptForSize]>;
755 let isAsmParserOnly = 1 in {
756 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
757 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
758 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
759 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
760 VR128:$src2))]>, XS, VEX_4V,
762 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
763 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
764 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
765 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
766 (load addr:$src2)))]>, XS, VEX_4V,
769 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
770 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
771 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
772 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
773 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
776 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
777 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
778 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
779 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
780 (load addr:$src2)))]>, XS,
784 def : Pat<(extloadf32 addr:$src),
785 (CVTSS2SDrr (MOVSSrm addr:$src))>,
786 Requires<[HasSSE2, OptForSpeed]>;
788 // Convert doubleword to packed single/double fp
789 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
790 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
791 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
792 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
793 TB, VEX, Requires<[HasAVX]>;
794 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
795 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
796 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
797 (bitconvert (memopv2i64 addr:$src))))]>,
798 TB, VEX, Requires<[HasAVX]>;
800 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
801 "cvtdq2ps\t{$src, $dst|$dst, $src}",
802 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
803 TB, Requires<[HasSSE2]>;
804 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
805 "cvtdq2ps\t{$src, $dst|$dst, $src}",
806 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
807 (bitconvert (memopv2i64 addr:$src))))]>,
808 TB, Requires<[HasSSE2]>;
810 // FIXME: why the non-intrinsic version is described as SSE3?
811 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
812 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
813 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
814 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
815 XS, VEX, Requires<[HasAVX]>;
816 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
817 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
818 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
819 (bitconvert (memopv2i64 addr:$src))))]>,
820 XS, VEX, Requires<[HasAVX]>;
822 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
823 "cvtdq2pd\t{$src, $dst|$dst, $src}",
824 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
825 XS, Requires<[HasSSE2]>;
826 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
827 "cvtdq2pd\t{$src, $dst|$dst, $src}",
828 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
829 (bitconvert (memopv2i64 addr:$src))))]>,
830 XS, Requires<[HasSSE2]>;
833 // Convert packed single/double fp to doubleword
834 let isAsmParserOnly = 1 in {
835 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
836 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
837 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
838 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
839 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
840 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
841 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
842 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
844 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
845 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
846 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
847 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
849 let isAsmParserOnly = 1 in {
850 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
851 "cvtps2dq\t{$src, $dst|$dst, $src}",
852 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
854 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
856 "cvtps2dq\t{$src, $dst|$dst, $src}",
857 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
858 (memop addr:$src)))]>, VEX;
860 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
861 "cvtps2dq\t{$src, $dst|$dst, $src}",
862 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
863 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
864 "cvtps2dq\t{$src, $dst|$dst, $src}",
865 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
866 (memop addr:$src)))]>;
868 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
869 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
870 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
871 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
872 XD, VEX, Requires<[HasAVX]>;
873 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
874 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
875 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
876 (memop addr:$src)))]>,
877 XD, VEX, Requires<[HasAVX]>;
879 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
880 "cvtpd2dq\t{$src, $dst|$dst, $src}",
881 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
882 XD, Requires<[HasSSE2]>;
883 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
884 "cvtpd2dq\t{$src, $dst|$dst, $src}",
885 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
886 (memop addr:$src)))]>,
887 XD, Requires<[HasSSE2]>;
890 // Convert with truncation packed single/double fp to doubleword
891 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
892 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
893 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
894 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
895 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
896 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
897 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
898 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
899 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
901 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
902 "cvttps2dq\t{$src, $dst|$dst, $src}",
904 (int_x86_sse2_cvttps2dq VR128:$src))]>;
905 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
906 "cvttps2dq\t{$src, $dst|$dst, $src}",
908 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
911 let isAsmParserOnly = 1 in {
912 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
913 "vcvttps2dq\t{$src, $dst|$dst, $src}",
915 (int_x86_sse2_cvttps2dq VR128:$src))]>,
916 XS, VEX, Requires<[HasAVX]>;
917 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
918 "vcvttps2dq\t{$src, $dst|$dst, $src}",
919 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
920 (memop addr:$src)))]>,
921 XS, VEX, Requires<[HasAVX]>;
924 let isAsmParserOnly = 1 in {
925 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
927 "cvttpd2dq\t{$src, $dst|$dst, $src}",
928 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
930 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
932 "cvttpd2dq\t{$src, $dst|$dst, $src}",
933 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
934 (memop addr:$src)))]>, VEX;
936 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
937 "cvttpd2dq\t{$src, $dst|$dst, $src}",
938 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
939 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
940 "cvttpd2dq\t{$src, $dst|$dst, $src}",
941 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
942 (memop addr:$src)))]>;
944 let isAsmParserOnly = 1 in {
945 // The assembler can recognize rr 256-bit instructions by seeing a ymm
946 // register, but the same isn't true when using memory operands instead.
947 // Provide other assembly rr and rm forms to address this explicitly.
948 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
949 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
950 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
951 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
954 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
955 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
957 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
960 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
961 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
963 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
966 // Convert packed single to packed double
967 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
968 // SSE2 instructions without OpSize prefix
969 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
970 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
971 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
972 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
973 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
974 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
975 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
976 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
978 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
979 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
980 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
981 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
983 let isAsmParserOnly = 1 in {
984 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
985 "vcvtps2pd\t{$src, $dst|$dst, $src}",
986 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
987 VEX, Requires<[HasAVX]>;
988 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
989 "vcvtps2pd\t{$src, $dst|$dst, $src}",
990 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
991 (load addr:$src)))]>,
992 VEX, Requires<[HasAVX]>;
994 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
995 "cvtps2pd\t{$src, $dst|$dst, $src}",
996 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
997 TB, Requires<[HasSSE2]>;
998 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
999 "cvtps2pd\t{$src, $dst|$dst, $src}",
1000 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1001 (load addr:$src)))]>,
1002 TB, Requires<[HasSSE2]>;
1004 // Convert packed double to packed single
1005 let isAsmParserOnly = 1 in {
1006 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1007 // register, but the same isn't true when using memory operands instead.
1008 // Provide other assembly rr and rm forms to address this explicitly.
1009 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1010 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1011 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1012 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1015 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1016 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1017 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1018 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1021 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1022 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1023 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1024 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1026 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1027 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1028 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1029 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1032 let isAsmParserOnly = 1 in {
1033 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1034 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1035 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1036 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1038 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1039 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1040 (memop addr:$src)))]>;
1042 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1043 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1044 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1045 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1046 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1047 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1048 (memop addr:$src)))]>;
1050 // AVX 256-bit register conversion intrinsics
1051 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1052 // whenever possible to avoid declaring two versions of each one.
1053 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1054 (VCVTDQ2PSYrr VR256:$src)>;
1055 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1056 (VCVTDQ2PSYrm addr:$src)>;
1058 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1059 (VCVTPD2PSYrr VR256:$src)>;
1060 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1061 (VCVTPD2PSYrm addr:$src)>;
1063 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1064 (VCVTPS2DQYrr VR256:$src)>;
1065 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1066 (VCVTPS2DQYrm addr:$src)>;
1068 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1069 (VCVTPS2PDYrr VR128:$src)>;
1070 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1071 (VCVTPS2PDYrm addr:$src)>;
1073 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1074 (VCVTTPD2DQYrr VR256:$src)>;
1075 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1076 (VCVTTPD2DQYrm addr:$src)>;
1078 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1079 (VCVTTPS2DQYrr VR256:$src)>;
1080 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1081 (VCVTTPS2DQYrm addr:$src)>;
1083 //===----------------------------------------------------------------------===//
1084 // SSE 1 & 2 - Compare Instructions
1085 //===----------------------------------------------------------------------===//
1087 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1088 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1089 string asm, string asm_alt> {
1090 def rr : SIi8<0xC2, MRMSrcReg,
1091 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1094 def rm : SIi8<0xC2, MRMSrcMem,
1095 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1097 // Accept explicit immediate argument form instead of comparison code.
1098 let isAsmParserOnly = 1 in {
1099 def rr_alt : SIi8<0xC2, MRMSrcReg,
1100 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1103 def rm_alt : SIi8<0xC2, MRMSrcMem,
1104 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1109 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1110 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1111 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1112 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1114 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1115 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1116 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1120 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1121 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1122 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1123 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1124 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1125 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1126 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1129 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1130 Intrinsic Int, string asm> {
1131 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1132 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1133 [(set VR128:$dst, (Int VR128:$src1,
1134 VR128:$src, imm:$cc))]>;
1135 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1136 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1137 [(set VR128:$dst, (Int VR128:$src1,
1138 (load addr:$src), imm:$cc))]>;
1141 // Aliases to match intrinsics which expect XMM operand(s).
1142 let isAsmParserOnly = 1 in {
1143 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1144 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1146 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1147 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1150 let Constraints = "$src1 = $dst" in {
1151 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1152 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1153 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1154 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1158 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1159 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1160 ValueType vt, X86MemOperand x86memop,
1161 PatFrag ld_frag, string OpcodeStr, Domain d> {
1162 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1163 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1164 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1165 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1166 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1167 [(set EFLAGS, (OpNode (vt RC:$src1),
1168 (ld_frag addr:$src2)))], d>;
1171 let Defs = [EFLAGS] in {
1172 let isAsmParserOnly = 1 in {
1173 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1174 "ucomiss", SSEPackedSingle>, VEX;
1175 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1176 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1177 let Pattern = []<dag> in {
1178 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1179 "comiss", SSEPackedSingle>, VEX;
1180 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1181 "comisd", SSEPackedDouble>, OpSize, VEX;
1184 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1185 load, "ucomiss", SSEPackedSingle>, VEX;
1186 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1187 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1189 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1190 load, "comiss", SSEPackedSingle>, VEX;
1191 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1192 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1194 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1195 "ucomiss", SSEPackedSingle>, TB;
1196 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1197 "ucomisd", SSEPackedDouble>, TB, OpSize;
1199 let Pattern = []<dag> in {
1200 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1201 "comiss", SSEPackedSingle>, TB;
1202 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1203 "comisd", SSEPackedDouble>, TB, OpSize;
1206 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1207 load, "ucomiss", SSEPackedSingle>, TB;
1208 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1209 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1211 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1212 "comiss", SSEPackedSingle>, TB;
1213 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1214 "comisd", SSEPackedDouble>, TB, OpSize;
1215 } // Defs = [EFLAGS]
1217 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1218 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1219 Intrinsic Int, string asm, string asm_alt,
1221 def rri : PIi8<0xC2, MRMSrcReg,
1222 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1223 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1224 def rmi : PIi8<0xC2, MRMSrcMem,
1225 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1226 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1227 // Accept explicit immediate argument form instead of comparison code.
1228 let isAsmParserOnly = 1 in {
1229 def rri_alt : PIi8<0xC2, MRMSrcReg,
1230 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1232 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1233 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1238 let isAsmParserOnly = 1 in {
1239 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1240 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1241 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1242 SSEPackedSingle>, VEX_4V;
1243 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1244 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1245 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1246 SSEPackedDouble>, OpSize, VEX_4V;
1247 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1248 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1249 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1250 SSEPackedSingle>, VEX_4V;
1251 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1252 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1253 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1254 SSEPackedDouble>, OpSize, VEX_4V;
1256 let Constraints = "$src1 = $dst" in {
1257 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1258 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1259 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1260 SSEPackedSingle>, TB;
1261 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1262 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1263 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1264 SSEPackedDouble>, TB, OpSize;
1267 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1268 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1269 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1270 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1271 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1272 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1273 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1274 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1276 //===----------------------------------------------------------------------===//
1277 // SSE 1 & 2 - Shuffle Instructions
1278 //===----------------------------------------------------------------------===//
1280 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1281 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1282 ValueType vt, string asm, PatFrag mem_frag,
1283 Domain d, bit IsConvertibleToThreeAddress = 0> {
1284 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1285 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1286 [(set RC:$dst, (vt (shufp:$src3
1287 RC:$src1, (mem_frag addr:$src2))))], d>;
1288 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1289 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1290 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1292 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1295 let isAsmParserOnly = 1 in {
1296 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1297 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1298 memopv4f32, SSEPackedSingle>, VEX_4V;
1299 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1300 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1301 memopv8f32, SSEPackedSingle>, VEX_4V;
1302 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1303 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1304 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1305 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1306 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1307 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1310 let Constraints = "$src1 = $dst" in {
1311 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1312 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1313 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1315 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1316 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1317 memopv2f64, SSEPackedDouble>, TB, OpSize;
1320 //===----------------------------------------------------------------------===//
1321 // SSE 1 & 2 - Unpack Instructions
1322 //===----------------------------------------------------------------------===//
1324 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1325 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1326 PatFrag mem_frag, RegisterClass RC,
1327 X86MemOperand x86memop, string asm,
1329 def rr : PI<opc, MRMSrcReg,
1330 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1332 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1333 def rm : PI<opc, MRMSrcMem,
1334 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1336 (vt (OpNode RC:$src1,
1337 (mem_frag addr:$src2))))], d>;
1340 let AddedComplexity = 10 in {
1341 let isAsmParserOnly = 1 in {
1342 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1343 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1344 SSEPackedSingle>, VEX_4V;
1345 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1346 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1347 SSEPackedDouble>, OpSize, VEX_4V;
1348 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1349 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1350 SSEPackedSingle>, VEX_4V;
1351 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1352 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1353 SSEPackedDouble>, OpSize, VEX_4V;
1355 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1356 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1357 SSEPackedSingle>, VEX_4V;
1358 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1359 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1360 SSEPackedDouble>, OpSize, VEX_4V;
1361 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1362 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1363 SSEPackedSingle>, VEX_4V;
1364 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1365 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1366 SSEPackedDouble>, OpSize, VEX_4V;
1369 let Constraints = "$src1 = $dst" in {
1370 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1371 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1372 SSEPackedSingle>, TB;
1373 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1374 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1375 SSEPackedDouble>, TB, OpSize;
1376 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1377 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1378 SSEPackedSingle>, TB;
1379 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1380 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1381 SSEPackedDouble>, TB, OpSize;
1382 } // Constraints = "$src1 = $dst"
1383 } // AddedComplexity
1385 //===----------------------------------------------------------------------===//
1386 // SSE 1 & 2 - Extract Floating-Point Sign mask
1387 //===----------------------------------------------------------------------===//
1389 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1390 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1392 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1393 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1394 [(set GR32:$dst, (Int RC:$src))], d>;
1395 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1396 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1400 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1401 SSEPackedSingle>, TB;
1402 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1403 SSEPackedDouble>, TB, OpSize;
1405 let isAsmParserOnly = 1 in {
1406 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1407 "movmskps", SSEPackedSingle>, VEX;
1408 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1409 "movmskpd", SSEPackedDouble>, OpSize,
1411 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1412 "movmskps", SSEPackedSingle>, VEX;
1413 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1414 "movmskpd", SSEPackedDouble>, OpSize,
1418 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1419 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1420 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1421 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1423 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1424 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1425 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1426 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1430 //===----------------------------------------------------------------------===//
1431 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1432 //===----------------------------------------------------------------------===//
1434 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1435 // names that start with 'Fs'.
1437 // Alias instructions that map fld0 to pxor for sse.
1438 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1439 canFoldAsLoad = 1 in {
1440 // FIXME: Set encoding to pseudo!
1441 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1442 [(set FR32:$dst, fp32imm0)]>,
1443 Requires<[HasSSE1]>, TB, OpSize;
1444 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1445 [(set FR64:$dst, fpimm0)]>,
1446 Requires<[HasSSE2]>, TB, OpSize;
1447 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1448 [(set FR32:$dst, fp32imm0)]>,
1449 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1450 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1451 [(set FR64:$dst, fpimm0)]>,
1452 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1455 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1456 // bits are disregarded.
1457 let neverHasSideEffects = 1 in {
1458 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1459 "movaps\t{$src, $dst|$dst, $src}", []>;
1460 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1461 "movapd\t{$src, $dst|$dst, $src}", []>;
1464 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1465 // bits are disregarded.
1466 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1467 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1468 "movaps\t{$src, $dst|$dst, $src}",
1469 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1470 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1471 "movapd\t{$src, $dst|$dst, $src}",
1472 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1475 //===----------------------------------------------------------------------===//
1476 // SSE 1 & 2 - Logical Instructions
1477 //===----------------------------------------------------------------------===//
1479 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1481 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1483 let isAsmParserOnly = 1 in {
1484 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1485 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1487 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1488 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1491 let Constraints = "$src1 = $dst" in {
1492 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1493 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1495 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1496 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1500 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1501 let mayLoad = 0 in {
1502 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1503 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1504 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1507 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1508 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1510 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1512 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1513 SDNode OpNode, int HasPat = 0,
1514 list<list<dag>> Pattern = []> {
1515 let isAsmParserOnly = 1, Pattern = []<dag> in {
1516 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1517 !strconcat(OpcodeStr, "ps"), f128mem,
1518 !if(HasPat, Pattern[0], // rr
1519 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1521 !if(HasPat, Pattern[2], // rm
1522 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1523 (memopv2i64 addr:$src2)))]), 0>,
1526 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1527 !strconcat(OpcodeStr, "pd"), f128mem,
1528 !if(HasPat, Pattern[1], // rr
1529 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1532 !if(HasPat, Pattern[3], // rm
1533 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1534 (memopv2i64 addr:$src2)))]), 0>,
1537 let Constraints = "$src1 = $dst" in {
1538 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1539 !strconcat(OpcodeStr, "ps"), f128mem,
1540 !if(HasPat, Pattern[0], // rr
1541 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1543 !if(HasPat, Pattern[2], // rm
1544 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1545 (memopv2i64 addr:$src2)))])>, TB;
1547 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1548 !strconcat(OpcodeStr, "pd"), f128mem,
1549 !if(HasPat, Pattern[1], // rr
1550 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1553 !if(HasPat, Pattern[3], // rm
1554 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1555 (memopv2i64 addr:$src2)))])>,
1560 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1562 let isAsmParserOnly = 1 in {
1563 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1564 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1565 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1567 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1568 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1572 // AVX 256-bit packed logical ops forms
1573 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1574 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1575 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1576 let isCommutable = 0 in
1577 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1579 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1580 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1581 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1582 let isCommutable = 0 in
1583 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1585 [(set VR128:$dst, (v2i64 (and (xor VR128:$src1,
1586 (bc_v2i64 (v4i32 immAllOnesV))),
1589 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1590 (bc_v2i64 (v2f64 VR128:$src2))))],
1592 [(set VR128:$dst, (v2i64 (and (xor (bc_v2i64 (v4f32 VR128:$src1)),
1593 (bc_v2i64 (v4i32 immAllOnesV))),
1594 (memopv2i64 addr:$src2))))],
1596 [(set VR128:$dst, (and (vnot (bc_v2i64 (v2f64 VR128:$src1))),
1597 (memopv2i64 addr:$src2)))]]>;
1599 //===----------------------------------------------------------------------===//
1600 // SSE 1 & 2 - Arithmetic Instructions
1601 //===----------------------------------------------------------------------===//
1603 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1606 /// In addition, we also have a special variant of the scalar form here to
1607 /// represent the associated intrinsic operation. This form is unlike the
1608 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1609 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1611 /// These three forms can each be reg+reg or reg+mem.
1614 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1616 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1618 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1619 OpNode, FR32, f32mem, Is2Addr>, XS;
1620 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1621 OpNode, FR64, f64mem, Is2Addr>, XD;
1624 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1626 let mayLoad = 0 in {
1627 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1628 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1629 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1630 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1634 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1636 let mayLoad = 0 in {
1637 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1638 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1639 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1640 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1644 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1646 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1647 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1648 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1649 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1652 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1654 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1655 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1656 SSEPackedSingle, Is2Addr>, TB;
1658 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1659 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1660 SSEPackedDouble, Is2Addr>, TB, OpSize;
1663 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1664 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1665 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1666 SSEPackedSingle, 0>, TB;
1668 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1669 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1670 SSEPackedDouble, 0>, TB, OpSize;
1673 // Binary Arithmetic instructions
1674 let isAsmParserOnly = 1 in {
1675 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1676 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1677 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1678 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1679 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1680 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1681 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1682 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1684 let isCommutable = 0 in {
1685 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1686 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1687 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1688 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1689 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1690 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1691 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1692 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1693 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1694 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1695 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1696 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1697 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1698 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1699 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1700 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1701 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1702 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1703 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1704 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1708 let Constraints = "$src1 = $dst" in {
1709 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1710 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1711 basic_sse12_fp_binop_s_int<0x58, "add">;
1712 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1713 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1714 basic_sse12_fp_binop_s_int<0x59, "mul">;
1716 let isCommutable = 0 in {
1717 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1718 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1719 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1720 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1721 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1722 basic_sse12_fp_binop_s_int<0x5E, "div">;
1723 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1724 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1725 basic_sse12_fp_binop_s_int<0x5F, "max">,
1726 basic_sse12_fp_binop_p_int<0x5F, "max">;
1727 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1728 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1729 basic_sse12_fp_binop_s_int<0x5D, "min">,
1730 basic_sse12_fp_binop_p_int<0x5D, "min">;
1735 /// In addition, we also have a special variant of the scalar form here to
1736 /// represent the associated intrinsic operation. This form is unlike the
1737 /// plain scalar form, in that it takes an entire vector (instead of a
1738 /// scalar) and leaves the top elements undefined.
1740 /// And, we have a special variant form for a full-vector intrinsic form.
1742 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1743 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1744 SDNode OpNode, Intrinsic F32Int> {
1745 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1746 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1747 [(set FR32:$dst, (OpNode FR32:$src))]>;
1748 // For scalar unary operations, fold a load into the operation
1749 // only in OptForSize mode. It eliminates an instruction, but it also
1750 // eliminates a whole-register clobber (the load), so it introduces a
1751 // partial register update condition.
1752 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1753 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1754 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1755 Requires<[HasSSE1, OptForSize]>;
1756 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1757 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1758 [(set VR128:$dst, (F32Int VR128:$src))]>;
1759 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1760 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1761 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1764 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1765 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1766 SDNode OpNode, Intrinsic F32Int> {
1767 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1768 !strconcat(OpcodeStr,
1769 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1770 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1771 !strconcat(OpcodeStr,
1772 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1773 []>, XS, Requires<[HasAVX, OptForSize]>;
1774 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1775 !strconcat(OpcodeStr,
1776 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1777 [(set VR128:$dst, (F32Int VR128:$src))]>;
1778 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1779 !strconcat(OpcodeStr,
1780 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1781 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1784 /// sse1_fp_unop_p - SSE1 unops in packed form.
1785 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1786 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1787 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1788 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1789 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1790 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1791 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1794 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1795 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1796 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1797 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1798 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1799 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1800 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1801 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1804 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1805 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1806 Intrinsic V4F32Int> {
1807 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1808 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1809 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1810 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1811 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1812 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1815 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1816 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1817 Intrinsic V4F32Int> {
1818 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1819 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1820 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1821 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1822 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1823 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1826 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1827 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1828 SDNode OpNode, Intrinsic F64Int> {
1829 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1830 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1831 [(set FR64:$dst, (OpNode FR64:$src))]>;
1832 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1833 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1834 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1835 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1836 Requires<[HasSSE2, OptForSize]>;
1837 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1838 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1839 [(set VR128:$dst, (F64Int VR128:$src))]>;
1840 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1841 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1842 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1845 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1846 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1847 SDNode OpNode, Intrinsic F64Int> {
1848 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1849 !strconcat(OpcodeStr,
1850 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1851 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1852 (ins FR64:$src1, f64mem:$src2),
1853 !strconcat(OpcodeStr,
1854 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1855 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1856 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1857 [(set VR128:$dst, (F64Int VR128:$src))]>;
1858 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1859 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1860 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1863 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1864 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1866 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1867 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1868 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1869 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1870 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1871 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1874 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1875 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1876 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1877 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1878 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1879 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1880 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1881 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1884 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1885 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1886 Intrinsic V2F64Int> {
1887 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1888 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1889 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1890 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1891 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1892 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1895 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1896 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1897 Intrinsic V2F64Int> {
1898 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1899 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1900 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1901 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1902 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1903 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1906 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1908 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1909 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1912 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1913 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1914 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1915 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1916 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1917 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1918 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1919 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1922 // Reciprocal approximations. Note that these typically require refinement
1923 // in order to obtain suitable precision.
1924 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1925 int_x86_sse_rsqrt_ss>, VEX_4V;
1926 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1927 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1928 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1929 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1931 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1933 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1934 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1935 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1936 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1940 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1941 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1942 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1943 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1944 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1945 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1947 // Reciprocal approximations. Note that these typically require refinement
1948 // in order to obtain suitable precision.
1949 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1950 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1951 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1952 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1953 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1954 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1956 // There is no f64 version of the reciprocal approximation instructions.
1958 //===----------------------------------------------------------------------===//
1959 // SSE 1 & 2 - Non-temporal stores
1960 //===----------------------------------------------------------------------===//
1962 let isAsmParserOnly = 1 in {
1963 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1964 (ins i128mem:$dst, VR128:$src),
1965 "movntps\t{$src, $dst|$dst, $src}",
1966 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1967 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1968 (ins i128mem:$dst, VR128:$src),
1969 "movntpd\t{$src, $dst|$dst, $src}",
1970 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1972 let ExeDomain = SSEPackedInt in
1973 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1974 (ins f128mem:$dst, VR128:$src),
1975 "movntdq\t{$src, $dst|$dst, $src}",
1976 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1978 let AddedComplexity = 400 in { // Prefer non-temporal versions
1979 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1980 (ins f128mem:$dst, VR128:$src),
1981 "movntps\t{$src, $dst|$dst, $src}",
1982 [(alignednontemporalstore (v4f32 VR128:$src),
1984 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1985 (ins f128mem:$dst, VR128:$src),
1986 "movntpd\t{$src, $dst|$dst, $src}",
1987 [(alignednontemporalstore (v2f64 VR128:$src),
1989 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1990 (ins f128mem:$dst, VR128:$src),
1991 "movntdq\t{$src, $dst|$dst, $src}",
1992 [(alignednontemporalstore (v2f64 VR128:$src),
1994 let ExeDomain = SSEPackedInt in
1995 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1996 (ins f128mem:$dst, VR128:$src),
1997 "movntdq\t{$src, $dst|$dst, $src}",
1998 [(alignednontemporalstore (v4f32 VR128:$src),
2001 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2002 (ins f256mem:$dst, VR256:$src),
2003 "movntps\t{$src, $dst|$dst, $src}",
2004 [(alignednontemporalstore (v8f32 VR256:$src),
2006 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2007 (ins f256mem:$dst, VR256:$src),
2008 "movntpd\t{$src, $dst|$dst, $src}",
2009 [(alignednontemporalstore (v4f64 VR256:$src),
2011 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2012 (ins f256mem:$dst, VR256:$src),
2013 "movntdq\t{$src, $dst|$dst, $src}",
2014 [(alignednontemporalstore (v4f64 VR256:$src),
2016 let ExeDomain = SSEPackedInt in
2017 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2018 (ins f256mem:$dst, VR256:$src),
2019 "movntdq\t{$src, $dst|$dst, $src}",
2020 [(alignednontemporalstore (v8f32 VR256:$src),
2025 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2026 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2027 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2028 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2029 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2030 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2032 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2033 "movntps\t{$src, $dst|$dst, $src}",
2034 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2035 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2036 "movntpd\t{$src, $dst|$dst, $src}",
2037 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2039 let ExeDomain = SSEPackedInt in
2040 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2041 "movntdq\t{$src, $dst|$dst, $src}",
2042 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2044 let AddedComplexity = 400 in { // Prefer non-temporal versions
2045 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2046 "movntps\t{$src, $dst|$dst, $src}",
2047 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2048 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2049 "movntpd\t{$src, $dst|$dst, $src}",
2050 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2052 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2053 "movntdq\t{$src, $dst|$dst, $src}",
2054 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2056 let ExeDomain = SSEPackedInt in
2057 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2058 "movntdq\t{$src, $dst|$dst, $src}",
2059 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2061 // There is no AVX form for instructions below this point
2062 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2063 "movnti\t{$src, $dst|$dst, $src}",
2064 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2065 TB, Requires<[HasSSE2]>;
2067 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2068 "movnti\t{$src, $dst|$dst, $src}",
2069 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2070 TB, Requires<[HasSSE2]>;
2073 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2074 "movnti\t{$src, $dst|$dst, $src}",
2075 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2076 TB, Requires<[HasSSE2]>;
2078 //===----------------------------------------------------------------------===//
2079 // SSE 1 & 2 - Misc Instructions (No AVX form)
2080 //===----------------------------------------------------------------------===//
2082 // Prefetch intrinsic.
2083 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2084 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2085 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2086 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2087 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2088 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2089 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2090 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2092 // Load, store, and memory fence
2093 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2094 TB, Requires<[HasSSE1]>;
2095 def : Pat<(X86SFence), (SFENCE)>;
2097 // Alias instructions that map zero vector to pxor / xorp* for sse.
2098 // We set canFoldAsLoad because this can be converted to a constant-pool
2099 // load of an all-zeros value if folding it would be beneficial.
2100 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2101 // JIT implementation, it does not expand the instructions below like
2102 // X86MCInstLower does.
2103 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2104 isCodeGenOnly = 1 in {
2105 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2106 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2107 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2108 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2109 let ExeDomain = SSEPackedInt in
2110 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2111 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2114 // The same as done above but for AVX. The 128-bit versions are the
2115 // same, but re-encoded. The 256-bit does not support PI version.
2116 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2117 // JIT implementatioan, it does not expand the instructions below like
2118 // X86MCInstLower does.
2119 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2120 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2121 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2122 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2123 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2124 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2125 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2126 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2127 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2128 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2129 let ExeDomain = SSEPackedInt in
2130 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2131 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2134 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2135 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2136 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2138 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2139 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2141 //===----------------------------------------------------------------------===//
2142 // SSE 1 & 2 - Load/Store XCSR register
2143 //===----------------------------------------------------------------------===//
2145 let isAsmParserOnly = 1 in {
2146 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2147 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2148 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2149 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2152 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2153 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2154 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2155 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2157 //===---------------------------------------------------------------------===//
2158 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2159 //===---------------------------------------------------------------------===//
2161 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2163 let isAsmParserOnly = 1 in {
2164 let neverHasSideEffects = 1 in {
2165 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2166 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2167 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2168 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2170 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2171 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2172 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2173 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2175 let canFoldAsLoad = 1, mayLoad = 1 in {
2176 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2177 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2178 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2179 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2180 let Predicates = [HasAVX] in {
2181 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2182 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2183 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2184 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2188 let mayStore = 1 in {
2189 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2190 (ins i128mem:$dst, VR128:$src),
2191 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2192 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2193 (ins i256mem:$dst, VR256:$src),
2194 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2195 let Predicates = [HasAVX] in {
2196 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2197 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2198 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2199 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2204 let neverHasSideEffects = 1 in
2205 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2206 "movdqa\t{$src, $dst|$dst, $src}", []>;
2208 let canFoldAsLoad = 1, mayLoad = 1 in {
2209 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2210 "movdqa\t{$src, $dst|$dst, $src}",
2211 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2212 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2213 "movdqu\t{$src, $dst|$dst, $src}",
2214 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2215 XS, Requires<[HasSSE2]>;
2218 let mayStore = 1 in {
2219 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2220 "movdqa\t{$src, $dst|$dst, $src}",
2221 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2222 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2223 "movdqu\t{$src, $dst|$dst, $src}",
2224 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2225 XS, Requires<[HasSSE2]>;
2228 // Intrinsic forms of MOVDQU load and store
2229 let isAsmParserOnly = 1 in {
2230 let canFoldAsLoad = 1 in
2231 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2232 "vmovdqu\t{$src, $dst|$dst, $src}",
2233 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2234 XS, VEX, Requires<[HasAVX]>;
2235 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2236 "vmovdqu\t{$src, $dst|$dst, $src}",
2237 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2238 XS, VEX, Requires<[HasAVX]>;
2241 let canFoldAsLoad = 1 in
2242 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2243 "movdqu\t{$src, $dst|$dst, $src}",
2244 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2245 XS, Requires<[HasSSE2]>;
2246 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2247 "movdqu\t{$src, $dst|$dst, $src}",
2248 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2249 XS, Requires<[HasSSE2]>;
2251 } // ExeDomain = SSEPackedInt
2253 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2254 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2255 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2257 //===---------------------------------------------------------------------===//
2258 // SSE2 - Packed Integer Arithmetic Instructions
2259 //===---------------------------------------------------------------------===//
2261 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2263 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2264 bit IsCommutable = 0, bit Is2Addr = 1> {
2265 let isCommutable = IsCommutable in
2266 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2267 (ins VR128:$src1, VR128:$src2),
2269 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2270 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2271 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2272 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2273 (ins VR128:$src1, i128mem:$src2),
2275 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2276 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2277 [(set VR128:$dst, (IntId VR128:$src1,
2278 (bitconvert (memopv2i64 addr:$src2))))]>;
2281 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2282 string OpcodeStr, Intrinsic IntId,
2283 Intrinsic IntId2, bit Is2Addr = 1> {
2284 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2285 (ins VR128:$src1, VR128:$src2),
2287 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2288 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2289 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2290 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2291 (ins VR128:$src1, i128mem:$src2),
2293 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2294 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2295 [(set VR128:$dst, (IntId VR128:$src1,
2296 (bitconvert (memopv2i64 addr:$src2))))]>;
2297 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2298 (ins VR128:$src1, i32i8imm:$src2),
2300 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2301 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2302 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2305 /// PDI_binop_rm - Simple SSE2 binary operator.
2306 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2307 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2308 let isCommutable = IsCommutable in
2309 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2310 (ins VR128:$src1, VR128:$src2),
2312 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2314 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2315 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2316 (ins VR128:$src1, i128mem:$src2),
2318 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2320 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2321 (bitconvert (memopv2i64 addr:$src2)))))]>;
2324 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2326 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2327 /// to collapse (bitconvert VT to VT) into its operand.
2329 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2330 bit IsCommutable = 0, bit Is2Addr = 1> {
2331 let isCommutable = IsCommutable in
2332 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2333 (ins VR128:$src1, VR128:$src2),
2335 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2337 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2338 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2339 (ins VR128:$src1, i128mem:$src2),
2341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2343 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2346 } // ExeDomain = SSEPackedInt
2348 // 128-bit Integer Arithmetic
2350 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2351 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2352 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2353 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2354 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2355 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2356 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2357 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2358 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2359 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2362 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2364 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2366 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2368 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2370 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2372 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2374 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2376 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2378 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2380 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2382 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2384 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2386 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2388 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2390 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2392 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2394 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2396 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2398 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2402 let Constraints = "$src1 = $dst" in {
2403 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2404 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2405 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2406 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2407 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2408 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2409 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2410 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2411 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2414 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2415 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2416 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2417 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2418 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2419 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2420 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2421 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2422 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2423 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2424 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2425 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2426 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2427 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2428 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2429 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2430 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2431 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2432 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2434 } // Constraints = "$src1 = $dst"
2436 //===---------------------------------------------------------------------===//
2437 // SSE2 - Packed Integer Logical Instructions
2438 //===---------------------------------------------------------------------===//
2440 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2441 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2442 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2444 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2445 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2447 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2448 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2451 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2452 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2454 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2455 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2457 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2458 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2461 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2462 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2464 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2465 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2468 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2469 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2470 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2472 let ExeDomain = SSEPackedInt in {
2473 let neverHasSideEffects = 1 in {
2474 // 128-bit logical shifts.
2475 def VPSLLDQri : PDIi8<0x73, MRM7r,
2476 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2477 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2479 def VPSRLDQri : PDIi8<0x73, MRM3r,
2480 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2481 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2483 // PSRADQri doesn't exist in SSE[1-3].
2485 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2486 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2487 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2488 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2489 VR128:$src2)))]>, VEX_4V;
2491 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2492 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2493 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2494 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2495 (memopv2i64 addr:$src2))))]>,
2500 let Constraints = "$src1 = $dst" in {
2501 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2502 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2503 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2504 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2505 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2506 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2508 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2509 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2510 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2511 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2512 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2513 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2515 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2516 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2517 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2518 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2520 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2521 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2522 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2524 let ExeDomain = SSEPackedInt in {
2525 let neverHasSideEffects = 1 in {
2526 // 128-bit logical shifts.
2527 def PSLLDQri : PDIi8<0x73, MRM7r,
2528 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2529 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2530 def PSRLDQri : PDIi8<0x73, MRM3r,
2531 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2532 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2533 // PSRADQri doesn't exist in SSE[1-3].
2535 def PANDNrr : PDI<0xDF, MRMSrcReg,
2536 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2537 "pandn\t{$src2, $dst|$dst, $src2}",
2538 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2541 def PANDNrm : PDI<0xDF, MRMSrcMem,
2542 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2543 "pandn\t{$src2, $dst|$dst, $src2}",
2544 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2545 (memopv2i64 addr:$src2))))]>;
2547 } // Constraints = "$src1 = $dst"
2549 let Predicates = [HasAVX] in {
2550 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2551 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2552 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2553 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2554 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2555 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2556 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2557 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2558 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2559 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2561 // Shift up / down and insert zero's.
2562 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2563 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2564 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2565 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2568 let Predicates = [HasSSE2] in {
2569 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2570 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2571 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2572 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2573 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2574 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2575 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2576 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2577 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2578 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2580 // Shift up / down and insert zero's.
2581 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2582 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2583 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2584 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2587 //===---------------------------------------------------------------------===//
2588 // SSE2 - Packed Integer Comparison Instructions
2589 //===---------------------------------------------------------------------===//
2591 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2592 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2594 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2596 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2598 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2600 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2602 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2606 let Constraints = "$src1 = $dst" in {
2607 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2608 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2609 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2610 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2611 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2612 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2613 } // Constraints = "$src1 = $dst"
2615 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2616 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2617 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2618 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2619 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2620 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2621 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2622 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2623 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2624 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2625 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2626 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2628 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2629 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2630 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2631 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2632 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2633 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2634 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2635 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2636 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2637 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2638 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2639 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2641 //===---------------------------------------------------------------------===//
2642 // SSE2 - Packed Integer Pack Instructions
2643 //===---------------------------------------------------------------------===//
2645 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2646 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2648 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2650 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2654 let Constraints = "$src1 = $dst" in {
2655 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2656 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2657 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2658 } // Constraints = "$src1 = $dst"
2660 //===---------------------------------------------------------------------===//
2661 // SSE2 - Packed Integer Shuffle Instructions
2662 //===---------------------------------------------------------------------===//
2664 let ExeDomain = SSEPackedInt in {
2665 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2667 def ri : Ii8<0x70, MRMSrcReg,
2668 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2669 !strconcat(OpcodeStr,
2670 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2671 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2673 def mi : Ii8<0x70, MRMSrcMem,
2674 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2675 !strconcat(OpcodeStr,
2676 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2677 [(set VR128:$dst, (vt (pshuf_frag:$src2
2678 (bc_frag (memopv2i64 addr:$src1)),
2681 } // ExeDomain = SSEPackedInt
2683 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2684 let AddedComplexity = 5 in
2685 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2688 // SSE2 with ImmT == Imm8 and XS prefix.
2689 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2692 // SSE2 with ImmT == Imm8 and XD prefix.
2693 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2697 let Predicates = [HasSSE2] in {
2698 let AddedComplexity = 5 in
2699 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2701 // SSE2 with ImmT == Imm8 and XS prefix.
2702 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2704 // SSE2 with ImmT == Imm8 and XD prefix.
2705 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2708 //===---------------------------------------------------------------------===//
2709 // SSE2 - Packed Integer Unpack Instructions
2710 //===---------------------------------------------------------------------===//
2712 let ExeDomain = SSEPackedInt in {
2713 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2714 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2715 def rr : PDI<opc, MRMSrcReg,
2716 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2718 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2719 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2720 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2721 def rm : PDI<opc, MRMSrcMem,
2722 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2724 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2725 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2726 [(set VR128:$dst, (unp_frag VR128:$src1,
2727 (bc_frag (memopv2i64
2731 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2732 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2734 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2736 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2739 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2740 /// knew to collapse (bitconvert VT to VT) into its operand.
2741 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2742 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2743 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2745 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2746 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2747 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2748 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2750 (v2i64 (unpckl VR128:$src1,
2751 (memopv2i64 addr:$src2))))]>, VEX_4V;
2753 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2755 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2757 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2760 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2761 /// knew to collapse (bitconvert VT to VT) into its operand.
2762 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2763 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2764 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2766 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2767 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2768 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2769 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2771 (v2i64 (unpckh VR128:$src1,
2772 (memopv2i64 addr:$src2))))]>, VEX_4V;
2775 let Constraints = "$src1 = $dst" in {
2776 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2777 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2778 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2780 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2781 /// knew to collapse (bitconvert VT to VT) into its operand.
2782 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2783 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2784 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2786 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2787 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2788 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2789 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2791 (v2i64 (unpckl VR128:$src1,
2792 (memopv2i64 addr:$src2))))]>;
2794 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2795 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2796 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2798 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2799 /// knew to collapse (bitconvert VT to VT) into its operand.
2800 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2801 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2802 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2804 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2805 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2806 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2807 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2809 (v2i64 (unpckh VR128:$src1,
2810 (memopv2i64 addr:$src2))))]>;
2813 } // ExeDomain = SSEPackedInt
2815 //===---------------------------------------------------------------------===//
2816 // SSE2 - Packed Integer Extract and Insert
2817 //===---------------------------------------------------------------------===//
2819 let ExeDomain = SSEPackedInt in {
2820 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2821 def rri : Ii8<0xC4, MRMSrcReg,
2822 (outs VR128:$dst), (ins VR128:$src1,
2823 GR32:$src2, i32i8imm:$src3),
2825 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2826 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2828 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2829 def rmi : Ii8<0xC4, MRMSrcMem,
2830 (outs VR128:$dst), (ins VR128:$src1,
2831 i16mem:$src2, i32i8imm:$src3),
2833 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2834 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2836 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
2841 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2842 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2843 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2844 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2845 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2846 imm:$src2))]>, OpSize, VEX;
2847 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2848 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2849 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2850 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2854 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2855 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2856 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2857 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2858 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2859 []>, OpSize, VEX_4V;
2862 let Constraints = "$src1 = $dst" in
2863 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2865 } // ExeDomain = SSEPackedInt
2867 //===---------------------------------------------------------------------===//
2868 // SSE2 - Packed Mask Creation
2869 //===---------------------------------------------------------------------===//
2871 let ExeDomain = SSEPackedInt in {
2873 let isAsmParserOnly = 1 in {
2874 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2875 "pmovmskb\t{$src, $dst|$dst, $src}",
2876 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2877 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2878 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2880 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2881 "pmovmskb\t{$src, $dst|$dst, $src}",
2882 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2884 } // ExeDomain = SSEPackedInt
2886 //===---------------------------------------------------------------------===//
2887 // SSE2 - Conditional Store
2888 //===---------------------------------------------------------------------===//
2890 let ExeDomain = SSEPackedInt in {
2892 let isAsmParserOnly = 1 in {
2894 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2895 (ins VR128:$src, VR128:$mask),
2896 "maskmovdqu\t{$mask, $src|$src, $mask}",
2897 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2899 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2900 (ins VR128:$src, VR128:$mask),
2901 "maskmovdqu\t{$mask, $src|$src, $mask}",
2902 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2906 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2907 "maskmovdqu\t{$mask, $src|$src, $mask}",
2908 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2910 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2911 "maskmovdqu\t{$mask, $src|$src, $mask}",
2912 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2914 } // ExeDomain = SSEPackedInt
2916 //===---------------------------------------------------------------------===//
2917 // SSE2 - Move Doubleword
2918 //===---------------------------------------------------------------------===//
2920 // Move Int Doubleword to Packed Double Int
2921 let isAsmParserOnly = 1 in {
2922 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2923 "movd\t{$src, $dst|$dst, $src}",
2925 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2926 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2927 "movd\t{$src, $dst|$dst, $src}",
2929 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2932 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2933 "movd\t{$src, $dst|$dst, $src}",
2935 (v4i32 (scalar_to_vector GR32:$src)))]>;
2936 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2937 "movd\t{$src, $dst|$dst, $src}",
2939 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2940 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2941 "mov{d|q}\t{$src, $dst|$dst, $src}",
2943 (v2i64 (scalar_to_vector GR64:$src)))]>;
2944 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2945 "mov{d|q}\t{$src, $dst|$dst, $src}",
2946 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2949 // Move Int Doubleword to Single Scalar
2950 let isAsmParserOnly = 1 in {
2951 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2952 "movd\t{$src, $dst|$dst, $src}",
2953 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2955 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2956 "movd\t{$src, $dst|$dst, $src}",
2957 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2960 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2961 "movd\t{$src, $dst|$dst, $src}",
2962 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2964 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2965 "movd\t{$src, $dst|$dst, $src}",
2966 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2968 // Move Packed Doubleword Int to Packed Double Int
2969 let isAsmParserOnly = 1 in {
2970 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2971 "movd\t{$src, $dst|$dst, $src}",
2972 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2974 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2975 (ins i32mem:$dst, VR128:$src),
2976 "movd\t{$src, $dst|$dst, $src}",
2977 [(store (i32 (vector_extract (v4i32 VR128:$src),
2978 (iPTR 0))), addr:$dst)]>, VEX;
2980 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2981 "movd\t{$src, $dst|$dst, $src}",
2982 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2984 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2985 "movd\t{$src, $dst|$dst, $src}",
2986 [(store (i32 (vector_extract (v4i32 VR128:$src),
2987 (iPTR 0))), addr:$dst)]>;
2989 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2990 "mov{d|q}\t{$src, $dst|$dst, $src}",
2991 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2993 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2994 "movq\t{$src, $dst|$dst, $src}",
2995 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2997 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2998 "mov{d|q}\t{$src, $dst|$dst, $src}",
2999 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3000 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3001 "movq\t{$src, $dst|$dst, $src}",
3002 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3004 // Move Scalar Single to Double Int
3005 let isAsmParserOnly = 1 in {
3006 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3007 "movd\t{$src, $dst|$dst, $src}",
3008 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3009 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3010 "movd\t{$src, $dst|$dst, $src}",
3011 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3013 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3014 "movd\t{$src, $dst|$dst, $src}",
3015 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3016 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3017 "movd\t{$src, $dst|$dst, $src}",
3018 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3020 // movd / movq to XMM register zero-extends
3021 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3022 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3023 "movd\t{$src, $dst|$dst, $src}",
3024 [(set VR128:$dst, (v4i32 (X86vzmovl
3025 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3027 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3028 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3029 [(set VR128:$dst, (v2i64 (X86vzmovl
3030 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3033 let AddedComplexity = 15 in {
3034 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3035 "movd\t{$src, $dst|$dst, $src}",
3036 [(set VR128:$dst, (v4i32 (X86vzmovl
3037 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3038 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3039 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3040 [(set VR128:$dst, (v2i64 (X86vzmovl
3041 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3044 let AddedComplexity = 20 in {
3045 let isAsmParserOnly = 1 in
3046 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3047 "movd\t{$src, $dst|$dst, $src}",
3049 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3050 (loadi32 addr:$src))))))]>,
3052 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3053 "movd\t{$src, $dst|$dst, $src}",
3055 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3056 (loadi32 addr:$src))))))]>;
3058 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3059 (MOVZDI2PDIrm addr:$src)>;
3060 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3061 (MOVZDI2PDIrm addr:$src)>;
3062 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3063 (MOVZDI2PDIrm addr:$src)>;
3066 //===---------------------------------------------------------------------===//
3067 // SSE2 - Move Quadword
3068 //===---------------------------------------------------------------------===//
3070 // Move Quadword Int to Packed Quadword Int
3071 let isAsmParserOnly = 1 in
3072 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3073 "vmovq\t{$src, $dst|$dst, $src}",
3075 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3076 VEX, Requires<[HasAVX]>;
3077 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3078 "movq\t{$src, $dst|$dst, $src}",
3080 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3081 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3083 // Move Packed Quadword Int to Quadword Int
3084 let isAsmParserOnly = 1 in
3085 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3086 "movq\t{$src, $dst|$dst, $src}",
3087 [(store (i64 (vector_extract (v2i64 VR128:$src),
3088 (iPTR 0))), addr:$dst)]>, VEX;
3089 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3090 "movq\t{$src, $dst|$dst, $src}",
3091 [(store (i64 (vector_extract (v2i64 VR128:$src),
3092 (iPTR 0))), addr:$dst)]>;
3094 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3095 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3097 // Store / copy lower 64-bits of a XMM register.
3098 let isAsmParserOnly = 1 in
3099 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3100 "movq\t{$src, $dst|$dst, $src}",
3101 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3102 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3103 "movq\t{$src, $dst|$dst, $src}",
3104 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3106 let AddedComplexity = 20, isAsmParserOnly = 1 in
3107 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3108 "vmovq\t{$src, $dst|$dst, $src}",
3110 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3111 (loadi64 addr:$src))))))]>,
3112 XS, VEX, Requires<[HasAVX]>;
3114 let AddedComplexity = 20 in {
3115 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3116 "movq\t{$src, $dst|$dst, $src}",
3118 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3119 (loadi64 addr:$src))))))]>,
3120 XS, Requires<[HasSSE2]>;
3122 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3123 (MOVZQI2PQIrm addr:$src)>;
3124 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3125 (MOVZQI2PQIrm addr:$src)>;
3126 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3129 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3130 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3131 let isAsmParserOnly = 1, AddedComplexity = 15 in
3132 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3133 "vmovq\t{$src, $dst|$dst, $src}",
3134 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3135 XS, VEX, Requires<[HasAVX]>;
3136 let AddedComplexity = 15 in
3137 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3138 "movq\t{$src, $dst|$dst, $src}",
3139 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3140 XS, Requires<[HasSSE2]>;
3142 let AddedComplexity = 20, isAsmParserOnly = 1 in
3143 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3144 "vmovq\t{$src, $dst|$dst, $src}",
3145 [(set VR128:$dst, (v2i64 (X86vzmovl
3146 (loadv2i64 addr:$src))))]>,
3147 XS, VEX, Requires<[HasAVX]>;
3148 let AddedComplexity = 20 in {
3149 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3150 "movq\t{$src, $dst|$dst, $src}",
3151 [(set VR128:$dst, (v2i64 (X86vzmovl
3152 (loadv2i64 addr:$src))))]>,
3153 XS, Requires<[HasSSE2]>;
3155 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3156 (MOVZPQILo2PQIrm addr:$src)>;
3159 // Instructions to match in the assembler
3160 let isAsmParserOnly = 1 in {
3161 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3162 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3163 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3164 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3165 // Recognize "movd" with GR64 destination, but encode as a "movq"
3166 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3167 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3170 // Instructions for the disassembler
3171 // xr = XMM register
3174 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3175 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3176 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3177 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3178 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3180 //===---------------------------------------------------------------------===//
3181 // SSE2 - Misc Instructions
3182 //===---------------------------------------------------------------------===//
3185 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3186 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3187 TB, Requires<[HasSSE2]>;
3189 // Load, store, and memory fence
3190 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3191 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3192 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3193 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3194 def : Pat<(X86LFence), (LFENCE)>;
3195 def : Pat<(X86MFence), (MFENCE)>;
3198 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3199 // was introduced with SSE2, it's backward compatible.
3200 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3202 // Alias instructions that map zero vector to pxor / xorp* for sse.
3203 // We set canFoldAsLoad because this can be converted to a constant-pool
3204 // load of an all-ones value if folding it would be beneficial.
3205 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3206 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3207 // FIXME: Change encoding to pseudo.
3208 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3209 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3211 //===---------------------------------------------------------------------===//
3212 // SSE3 - Conversion Instructions
3213 //===---------------------------------------------------------------------===//
3215 // Convert Packed Double FP to Packed DW Integers
3216 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3217 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3218 // register, but the same isn't true when using memory operands instead.
3219 // Provide other assembly rr and rm forms to address this explicitly.
3220 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3221 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3222 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3223 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3226 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3227 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3228 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3229 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3232 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3233 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3234 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3235 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3238 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3239 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3240 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3241 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3243 // Convert Packed DW Integers to Packed Double FP
3244 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3245 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3246 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3247 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3248 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3249 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3250 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3251 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3252 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3255 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3256 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3257 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3258 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3260 // AVX 256-bit register conversion intrinsics
3261 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3262 (VCVTDQ2PDYrr VR128:$src)>;
3263 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3264 (VCVTDQ2PDYrm addr:$src)>;
3266 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3267 (VCVTPD2DQYrr VR256:$src)>;
3268 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3269 (VCVTPD2DQYrm addr:$src)>;
3271 //===---------------------------------------------------------------------===//
3272 // SSE3 - Move Instructions
3273 //===---------------------------------------------------------------------===//
3275 // Replicate Single FP
3276 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3277 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3278 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3279 [(set VR128:$dst, (v4f32 (rep_frag
3280 VR128:$src, (undef))))]>;
3281 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3282 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3283 [(set VR128:$dst, (rep_frag
3284 (memopv4f32 addr:$src), (undef)))]>;
3287 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3289 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3290 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3291 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3292 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3295 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3296 // FIXME: Merge above classes when we have patterns for the ymm version
3297 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3298 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3299 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3300 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3302 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3303 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3305 // Replicate Double FP
3306 multiclass sse3_replicate_dfp<string OpcodeStr> {
3307 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3308 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3309 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3310 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3311 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3313 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3317 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3318 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3319 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3321 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3322 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3326 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3327 // FIXME: Merge above classes when we have patterns for the ymm version
3328 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3329 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3331 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3333 // Move Unaligned Integer
3334 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3335 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3336 "vlddqu\t{$src, $dst|$dst, $src}",
3337 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3338 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3339 "vlddqu\t{$src, $dst|$dst, $src}",
3340 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3342 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3343 "lddqu\t{$src, $dst|$dst, $src}",
3344 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3346 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3348 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3350 // Several Move patterns
3351 let AddedComplexity = 5 in {
3352 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3353 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3354 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3355 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3356 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3357 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3358 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3359 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3362 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3363 let AddedComplexity = 15 in
3364 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3365 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3366 let AddedComplexity = 20 in
3367 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3368 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3370 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3371 let AddedComplexity = 15 in
3372 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3373 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3374 let AddedComplexity = 20 in
3375 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3376 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3378 //===---------------------------------------------------------------------===//
3379 // SSE3 - Arithmetic
3380 //===---------------------------------------------------------------------===//
3382 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3383 X86MemOperand x86memop, bit Is2Addr = 1> {
3384 def rr : I<0xD0, MRMSrcReg,
3385 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3387 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3388 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3389 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3390 def rm : I<0xD0, MRMSrcMem,
3391 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3393 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3394 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3395 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3398 let isAsmParserOnly = 1, Predicates = [HasAVX],
3399 ExeDomain = SSEPackedDouble in {
3400 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3401 f128mem, 0>, XD, VEX_4V;
3402 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3403 f128mem, 0>, OpSize, VEX_4V;
3404 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3405 f256mem, 0>, XD, VEX_4V;
3406 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3407 f256mem, 0>, OpSize, VEX_4V;
3409 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3410 ExeDomain = SSEPackedDouble in {
3411 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3413 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3414 f128mem>, TB, OpSize;
3417 //===---------------------------------------------------------------------===//
3418 // SSE3 Instructions
3419 //===---------------------------------------------------------------------===//
3422 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3423 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3424 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3426 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3427 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3428 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3430 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3432 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3433 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3434 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3436 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3437 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3438 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3440 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3441 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3442 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3444 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3446 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3447 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3448 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3451 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3452 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3453 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3454 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3455 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3456 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3457 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3458 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3459 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3460 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3461 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3462 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3463 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3464 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3465 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3466 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3467 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3470 let Constraints = "$src1 = $dst" in {
3471 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3472 int_x86_sse3_hadd_ps>;
3473 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3474 int_x86_sse3_hadd_pd>;
3475 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3476 int_x86_sse3_hsub_ps>;
3477 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3478 int_x86_sse3_hsub_pd>;
3481 //===---------------------------------------------------------------------===//
3482 // SSSE3 - Packed Absolute Instructions
3483 //===---------------------------------------------------------------------===//
3486 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3487 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3488 PatFrag mem_frag128, Intrinsic IntId128> {
3489 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3491 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3492 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3495 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3497 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3500 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3503 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3504 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3505 int_x86_ssse3_pabs_b_128>, VEX;
3506 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3507 int_x86_ssse3_pabs_w_128>, VEX;
3508 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3509 int_x86_ssse3_pabs_d_128>, VEX;
3512 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3513 int_x86_ssse3_pabs_b_128>;
3514 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3515 int_x86_ssse3_pabs_w_128>;
3516 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3517 int_x86_ssse3_pabs_d_128>;
3519 //===---------------------------------------------------------------------===//
3520 // SSSE3 - Packed Binary Operator Instructions
3521 //===---------------------------------------------------------------------===//
3523 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3524 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3525 PatFrag mem_frag128, Intrinsic IntId128,
3527 let isCommutable = 1 in
3528 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3529 (ins VR128:$src1, VR128:$src2),
3531 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3532 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3533 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3535 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3536 (ins VR128:$src1, i128mem:$src2),
3538 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3539 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3541 (IntId128 VR128:$src1,
3542 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3545 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3546 let isCommutable = 0 in {
3547 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3548 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3549 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3550 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3551 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3552 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3553 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3554 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3555 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3556 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3557 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3558 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3559 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3560 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3561 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3562 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3563 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3564 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3565 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3566 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3567 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3568 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3570 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3571 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3574 // None of these have i8 immediate fields.
3575 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3576 let isCommutable = 0 in {
3577 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3578 int_x86_ssse3_phadd_w_128>;
3579 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3580 int_x86_ssse3_phadd_d_128>;
3581 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3582 int_x86_ssse3_phadd_sw_128>;
3583 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3584 int_x86_ssse3_phsub_w_128>;
3585 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3586 int_x86_ssse3_phsub_d_128>;
3587 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3588 int_x86_ssse3_phsub_sw_128>;
3589 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3590 int_x86_ssse3_pmadd_ub_sw_128>;
3591 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3592 int_x86_ssse3_pshuf_b_128>;
3593 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3594 int_x86_ssse3_psign_b_128>;
3595 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3596 int_x86_ssse3_psign_w_128>;
3597 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3598 int_x86_ssse3_psign_d_128>;
3600 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3601 int_x86_ssse3_pmul_hr_sw_128>;
3604 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3605 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3606 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3607 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3609 //===---------------------------------------------------------------------===//
3610 // SSSE3 - Packed Align Instruction Patterns
3611 //===---------------------------------------------------------------------===//
3613 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3614 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3615 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3617 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3619 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3621 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3622 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3624 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3626 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3630 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3631 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3632 let Constraints = "$src1 = $dst" in
3633 defm PALIGN : ssse3_palign<"palignr">;
3635 let AddedComplexity = 5 in {
3636 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3637 (PALIGNR128rr VR128:$src2, VR128:$src1,
3638 (SHUFFLE_get_palign_imm VR128:$src3))>,
3639 Requires<[HasSSSE3]>;
3640 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3641 (PALIGNR128rr VR128:$src2, VR128:$src1,
3642 (SHUFFLE_get_palign_imm VR128:$src3))>,
3643 Requires<[HasSSSE3]>;
3644 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3645 (PALIGNR128rr VR128:$src2, VR128:$src1,
3646 (SHUFFLE_get_palign_imm VR128:$src3))>,
3647 Requires<[HasSSSE3]>;
3648 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3649 (PALIGNR128rr VR128:$src2, VR128:$src1,
3650 (SHUFFLE_get_palign_imm VR128:$src3))>,
3651 Requires<[HasSSSE3]>;
3654 //===---------------------------------------------------------------------===//
3655 // SSSE3 Misc Instructions
3656 //===---------------------------------------------------------------------===//
3658 // Thread synchronization
3659 let usesCustomInserter = 1 in {
3660 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3661 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3662 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3663 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3666 let Uses = [EAX, ECX, EDX] in
3667 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3668 Requires<[HasSSE3]>;
3669 let Uses = [ECX, EAX] in
3670 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3671 Requires<[HasSSE3]>;
3673 //===---------------------------------------------------------------------===//
3674 // Non-Instruction Patterns
3675 //===---------------------------------------------------------------------===//
3677 // extload f32 -> f64. This matches load+fextend because we have a hack in
3678 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3680 // Since these loads aren't folded into the fextend, we have to match it
3682 let Predicates = [HasSSE2] in
3683 def : Pat<(fextend (loadf32 addr:$src)),
3684 (CVTSS2SDrm addr:$src)>;
3687 let Predicates = [HasXMMInt] in {
3688 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3689 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3690 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3691 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3692 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3693 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3694 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3695 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3696 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3697 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3698 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3699 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3700 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3701 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3702 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3703 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3704 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3705 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3706 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3707 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3708 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3709 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3710 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3711 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3712 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3713 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3714 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3715 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3716 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3717 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3720 let Predicates = [HasAVX] in {
3721 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3724 // Move scalar to XMM zero-extended
3725 // movd to XMM register zero-extends
3726 let AddedComplexity = 15 in {
3727 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3728 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3729 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3730 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3731 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3732 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3733 (MOVSSrr (v4f32 (V_SET0PS)),
3734 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3735 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3736 (MOVSSrr (v4i32 (V_SET0PI)),
3737 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3740 // Splat v2f64 / v2i64
3741 let AddedComplexity = 10 in {
3742 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3743 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3744 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3745 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3746 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3747 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3748 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3749 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3752 // Special unary SHUFPSrri case.
3753 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3754 (SHUFPSrri VR128:$src1, VR128:$src1,
3755 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3756 let AddedComplexity = 5 in
3757 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3758 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3759 Requires<[HasSSE2]>;
3760 // Special unary SHUFPDrri case.
3761 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3762 (SHUFPDrri VR128:$src1, VR128:$src1,
3763 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3764 Requires<[HasSSE2]>;
3765 // Special unary SHUFPDrri case.
3766 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3767 (SHUFPDrri VR128:$src1, VR128:$src1,
3768 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3769 Requires<[HasSSE2]>;
3770 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3771 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3772 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3773 Requires<[HasSSE2]>;
3775 // Special binary v4i32 shuffle cases with SHUFPS.
3776 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3777 (SHUFPSrri VR128:$src1, VR128:$src2,
3778 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3779 Requires<[HasSSE2]>;
3780 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3781 (SHUFPSrmi VR128:$src1, addr:$src2,
3782 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3783 Requires<[HasSSE2]>;
3784 // Special binary v2i64 shuffle cases using SHUFPDrri.
3785 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3786 (SHUFPDrri VR128:$src1, VR128:$src2,
3787 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3788 Requires<[HasSSE2]>;
3790 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3791 let AddedComplexity = 15 in {
3792 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3793 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3794 Requires<[OptForSpeed, HasSSE2]>;
3795 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3796 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3797 Requires<[OptForSpeed, HasSSE2]>;
3799 let AddedComplexity = 10 in {
3800 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3801 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3802 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3803 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3804 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3805 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3806 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3807 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3810 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3811 let AddedComplexity = 15 in {
3812 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3813 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3814 Requires<[OptForSpeed, HasSSE2]>;
3815 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3816 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3817 Requires<[OptForSpeed, HasSSE2]>;
3819 let AddedComplexity = 10 in {
3820 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3821 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3822 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3823 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3824 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3825 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3826 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3827 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3830 let AddedComplexity = 20 in {
3831 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3832 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3833 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3835 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3836 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3837 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3839 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3840 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3841 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3842 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3843 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3846 let AddedComplexity = 20 in {
3847 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3848 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3849 (MOVLPSrm VR128:$src1, addr:$src2)>;
3850 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3851 (MOVLPDrm VR128:$src1, addr:$src2)>;
3852 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3853 (MOVLPSrm VR128:$src1, addr:$src2)>;
3854 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3855 (MOVLPDrm VR128:$src1, addr:$src2)>;
3858 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3859 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3860 (MOVLPSmr addr:$src1, VR128:$src2)>;
3861 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3862 (MOVLPDmr addr:$src1, VR128:$src2)>;
3863 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3865 (MOVLPSmr addr:$src1, VR128:$src2)>;
3866 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3867 (MOVLPDmr addr:$src1, VR128:$src2)>;
3869 let AddedComplexity = 15 in {
3870 // Setting the lowest element in the vector.
3871 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3872 (MOVSSrr (v4i32 VR128:$src1),
3873 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3874 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3875 (MOVSDrr (v2i64 VR128:$src1),
3876 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3878 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3879 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3880 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3881 Requires<[HasSSE2]>;
3882 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3883 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3884 Requires<[HasSSE2]>;
3887 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3888 // fall back to this for SSE1)
3889 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3890 (SHUFPSrri VR128:$src2, VR128:$src1,
3891 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3893 // Set lowest element and zero upper elements.
3894 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3895 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3897 // Some special case pandn patterns.
3898 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3900 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3901 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3903 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3904 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3906 (PANDNrr VR128:$src1, VR128:$src2)>, Requires<[HasSSE2]>;
3908 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v4i32 immAllOnesV))),
3909 (memop addr:$src2))),
3910 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3911 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v8i16 immAllOnesV))),
3912 (memop addr:$src2))),
3913 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3914 def : Pat<(v2i64 (and (xor VR128:$src1, (bc_v2i64 (v16i8 immAllOnesV))),
3915 (memop addr:$src2))),
3916 (PANDNrm VR128:$src1, addr:$src2)>, Requires<[HasSSE2]>;
3918 // vector -> vector casts
3919 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3920 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3921 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3922 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3924 // Use movaps / movups for SSE integer load / store (one byte shorter).
3925 let Predicates = [HasSSE1] in {
3926 def : Pat<(alignedloadv4i32 addr:$src),
3927 (MOVAPSrm addr:$src)>;
3928 def : Pat<(loadv4i32 addr:$src),
3929 (MOVUPSrm addr:$src)>;
3930 def : Pat<(alignedloadv2i64 addr:$src),
3931 (MOVAPSrm addr:$src)>;
3932 def : Pat<(loadv2i64 addr:$src),
3933 (MOVUPSrm addr:$src)>;
3935 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3936 (MOVAPSmr addr:$dst, VR128:$src)>;
3937 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3938 (MOVAPSmr addr:$dst, VR128:$src)>;
3939 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3940 (MOVAPSmr addr:$dst, VR128:$src)>;
3941 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3942 (MOVAPSmr addr:$dst, VR128:$src)>;
3943 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3944 (MOVUPSmr addr:$dst, VR128:$src)>;
3945 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3946 (MOVUPSmr addr:$dst, VR128:$src)>;
3947 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3948 (MOVUPSmr addr:$dst, VR128:$src)>;
3949 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3950 (MOVUPSmr addr:$dst, VR128:$src)>;
3953 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3954 let Predicates = [HasAVX] in {
3955 def : Pat<(alignedloadv4i32 addr:$src),
3956 (VMOVAPSrm addr:$src)>;
3957 def : Pat<(loadv4i32 addr:$src),
3958 (VMOVUPSrm addr:$src)>;
3959 def : Pat<(alignedloadv2i64 addr:$src),
3960 (VMOVAPSrm addr:$src)>;
3961 def : Pat<(loadv2i64 addr:$src),
3962 (VMOVUPSrm addr:$src)>;
3964 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3965 (VMOVAPSmr addr:$dst, VR128:$src)>;
3966 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3967 (VMOVAPSmr addr:$dst, VR128:$src)>;
3968 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3969 (VMOVAPSmr addr:$dst, VR128:$src)>;
3970 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3971 (VMOVAPSmr addr:$dst, VR128:$src)>;
3972 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3973 (VMOVUPSmr addr:$dst, VR128:$src)>;
3974 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3975 (VMOVUPSmr addr:$dst, VR128:$src)>;
3976 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3977 (VMOVUPSmr addr:$dst, VR128:$src)>;
3978 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3979 (VMOVUPSmr addr:$dst, VR128:$src)>;
3982 //===----------------------------------------------------------------------===//
3983 // SSE4.1 - Packed Move with Sign/Zero Extend
3984 //===----------------------------------------------------------------------===//
3986 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3987 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3988 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3989 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3991 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3992 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3994 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3998 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3999 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4001 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4003 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4005 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4007 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4009 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4013 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4014 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4015 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4016 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4017 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4018 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4020 // Common patterns involving scalar load.
4021 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4022 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4023 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4024 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4026 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4027 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4028 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4029 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4031 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4032 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4033 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4034 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4036 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4037 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4038 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4039 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4041 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4042 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4043 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4044 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4046 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4047 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4048 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4049 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4052 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4053 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4054 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4055 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4057 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4058 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4060 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4064 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4065 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4067 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4069 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4071 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4075 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4076 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4077 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4078 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4080 // Common patterns involving scalar load
4081 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4082 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4083 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4084 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4086 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4087 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4088 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4089 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4092 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4093 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4094 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4095 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4097 // Expecting a i16 load any extended to i32 value.
4098 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4099 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4100 [(set VR128:$dst, (IntId (bitconvert
4101 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4105 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4106 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4108 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4111 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4112 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4114 // Common patterns involving scalar load
4115 def : Pat<(int_x86_sse41_pmovsxbq
4116 (bitconvert (v4i32 (X86vzmovl
4117 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4118 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4120 def : Pat<(int_x86_sse41_pmovzxbq
4121 (bitconvert (v4i32 (X86vzmovl
4122 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4123 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4125 //===----------------------------------------------------------------------===//
4126 // SSE4.1 - Extract Instructions
4127 //===----------------------------------------------------------------------===//
4129 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4130 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4131 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4132 (ins VR128:$src1, i32i8imm:$src2),
4133 !strconcat(OpcodeStr,
4134 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4135 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4137 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4138 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4139 !strconcat(OpcodeStr,
4140 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4143 // There's an AssertZext in the way of writing the store pattern
4144 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4147 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4148 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4149 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4150 (ins VR128:$src1, i32i8imm:$src2),
4151 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4154 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4157 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4158 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4159 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4160 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4161 !strconcat(OpcodeStr,
4162 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4165 // There's an AssertZext in the way of writing the store pattern
4166 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4169 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4170 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4172 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4175 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4176 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4177 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4178 (ins VR128:$src1, i32i8imm:$src2),
4179 !strconcat(OpcodeStr,
4180 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4182 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4183 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4184 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4185 !strconcat(OpcodeStr,
4186 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4187 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4188 addr:$dst)]>, OpSize;
4191 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4192 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4194 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4196 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4197 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4198 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4199 (ins VR128:$src1, i32i8imm:$src2),
4200 !strconcat(OpcodeStr,
4201 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4203 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4204 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4205 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4206 !strconcat(OpcodeStr,
4207 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4208 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4209 addr:$dst)]>, OpSize, REX_W;
4212 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4213 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4215 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4217 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4219 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4220 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4221 (ins VR128:$src1, i32i8imm:$src2),
4222 !strconcat(OpcodeStr,
4223 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4225 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4227 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4228 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4229 !strconcat(OpcodeStr,
4230 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4231 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4232 addr:$dst)]>, OpSize;
4235 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4236 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4237 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4238 (ins VR128:$src1, i32i8imm:$src2),
4239 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4242 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4244 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4245 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4248 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4249 Requires<[HasSSE41]>;
4251 //===----------------------------------------------------------------------===//
4252 // SSE4.1 - Insert Instructions
4253 //===----------------------------------------------------------------------===//
4255 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4256 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4257 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4259 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4261 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4263 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4264 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4265 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4267 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4269 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4271 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4272 imm:$src3))]>, OpSize;
4275 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4276 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4277 let Constraints = "$src1 = $dst" in
4278 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4280 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4281 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4282 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4284 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4286 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4288 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4290 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4291 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4293 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4295 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4297 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4298 imm:$src3)))]>, OpSize;
4301 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4302 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4303 let Constraints = "$src1 = $dst" in
4304 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4306 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4307 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4308 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4310 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4312 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4314 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4316 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4317 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4319 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4321 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4323 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4324 imm:$src3)))]>, OpSize;
4327 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4328 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4329 let Constraints = "$src1 = $dst" in
4330 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4332 // insertps has a few different modes, there's the first two here below which
4333 // are optimized inserts that won't zero arbitrary elements in the destination
4334 // vector. The next one matches the intrinsic and could zero arbitrary elements
4335 // in the target vector.
4336 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4337 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4338 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4340 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4342 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4344 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4346 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4347 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4349 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4351 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4353 (X86insrtps VR128:$src1,
4354 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4355 imm:$src3))]>, OpSize;
4358 let Constraints = "$src1 = $dst" in
4359 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4360 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4361 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4363 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4364 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4366 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4367 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4368 Requires<[HasSSE41]>;
4370 //===----------------------------------------------------------------------===//
4371 // SSE4.1 - Round Instructions
4372 //===----------------------------------------------------------------------===//
4374 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4375 X86MemOperand x86memop, RegisterClass RC,
4376 PatFrag mem_frag32, PatFrag mem_frag64,
4377 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4378 // Intrinsic operation, reg.
4379 // Vector intrinsic operation, reg
4380 def PSr : SS4AIi8<opcps, MRMSrcReg,
4381 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4382 !strconcat(OpcodeStr,
4383 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4384 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4387 // Vector intrinsic operation, mem
4388 def PSm : Ii8<opcps, MRMSrcMem,
4389 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4390 !strconcat(OpcodeStr,
4391 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4393 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4395 Requires<[HasSSE41]>;
4397 // Vector intrinsic operation, reg
4398 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4399 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4400 !strconcat(OpcodeStr,
4401 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4402 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4405 // Vector intrinsic operation, mem
4406 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4407 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4408 !strconcat(OpcodeStr,
4409 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4411 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4415 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4416 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4417 // Intrinsic operation, reg.
4418 // Vector intrinsic operation, reg
4419 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4420 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4421 !strconcat(OpcodeStr,
4422 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4425 // Vector intrinsic operation, mem
4426 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4427 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4428 !strconcat(OpcodeStr,
4429 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4430 []>, TA, OpSize, Requires<[HasSSE41]>;
4432 // Vector intrinsic operation, reg
4433 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4434 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4435 !strconcat(OpcodeStr,
4436 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4439 // Vector intrinsic operation, mem
4440 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4441 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4442 !strconcat(OpcodeStr,
4443 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4447 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4450 Intrinsic F64Int, bit Is2Addr = 1> {
4451 // Intrinsic operation, reg.
4452 def SSr : SS4AIi8<opcss, MRMSrcReg,
4453 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4455 !strconcat(OpcodeStr,
4456 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4457 !strconcat(OpcodeStr,
4458 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4459 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4462 // Intrinsic operation, mem.
4463 def SSm : SS4AIi8<opcss, MRMSrcMem,
4464 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4466 !strconcat(OpcodeStr,
4467 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4468 !strconcat(OpcodeStr,
4469 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4471 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4474 // Intrinsic operation, reg.
4475 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4476 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4478 !strconcat(OpcodeStr,
4479 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4480 !strconcat(OpcodeStr,
4481 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4482 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4485 // Intrinsic operation, mem.
4486 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4487 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4489 !strconcat(OpcodeStr,
4490 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4491 !strconcat(OpcodeStr,
4492 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4494 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4498 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4500 // Intrinsic operation, reg.
4501 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4502 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4503 !strconcat(OpcodeStr,
4504 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4507 // Intrinsic operation, mem.
4508 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4509 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4510 !strconcat(OpcodeStr,
4511 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4514 // Intrinsic operation, reg.
4515 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4516 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4517 !strconcat(OpcodeStr,
4518 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4521 // Intrinsic operation, mem.
4522 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4523 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4524 !strconcat(OpcodeStr,
4525 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4529 // FP round - roundss, roundps, roundsd, roundpd
4530 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4532 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4533 memopv4f32, memopv2f64,
4534 int_x86_sse41_round_ps,
4535 int_x86_sse41_round_pd>, VEX;
4536 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4537 memopv8f32, memopv4f64,
4538 int_x86_avx_round_ps_256,
4539 int_x86_avx_round_pd_256>, VEX;
4540 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4541 int_x86_sse41_round_ss,
4542 int_x86_sse41_round_sd, 0>, VEX_4V;
4544 // Instructions for the assembler
4545 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4547 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4549 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4552 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4553 memopv4f32, memopv2f64,
4554 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4555 let Constraints = "$src1 = $dst" in
4556 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4557 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4559 //===----------------------------------------------------------------------===//
4560 // SSE4.1 - Packed Bit Test
4561 //===----------------------------------------------------------------------===//
4563 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4564 // the intel intrinsic that corresponds to this.
4565 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4566 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4567 "vptest\t{$src2, $src1|$src1, $src2}",
4568 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4570 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4571 "vptest\t{$src2, $src1|$src1, $src2}",
4572 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4575 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4576 "vptest\t{$src2, $src1|$src1, $src2}",
4577 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4579 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4580 "vptest\t{$src2, $src1|$src1, $src2}",
4581 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4585 let Defs = [EFLAGS] in {
4586 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4587 "ptest \t{$src2, $src1|$src1, $src2}",
4588 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4590 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4591 "ptest \t{$src2, $src1|$src1, $src2}",
4592 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4596 // The bit test instructions below are AVX only
4597 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4598 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4599 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4600 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4601 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4602 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4603 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4604 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4608 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4609 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4610 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4611 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4612 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4615 //===----------------------------------------------------------------------===//
4616 // SSE4.1 - Misc Instructions
4617 //===----------------------------------------------------------------------===//
4619 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4620 "popcnt{w}\t{$src, $dst|$dst, $src}",
4621 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4622 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4623 "popcnt{w}\t{$src, $dst|$dst, $src}",
4624 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4626 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4627 "popcnt{l}\t{$src, $dst|$dst, $src}",
4628 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4629 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4630 "popcnt{l}\t{$src, $dst|$dst, $src}",
4631 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4633 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4634 "popcnt{q}\t{$src, $dst|$dst, $src}",
4635 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4636 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4637 "popcnt{q}\t{$src, $dst|$dst, $src}",
4638 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4642 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4643 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4644 Intrinsic IntId128> {
4645 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4647 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4648 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4649 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4651 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4654 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4657 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4658 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4659 int_x86_sse41_phminposuw>, VEX;
4660 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4661 int_x86_sse41_phminposuw>;
4663 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4664 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4665 Intrinsic IntId128, bit Is2Addr = 1> {
4666 let isCommutable = 1 in
4667 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4668 (ins VR128:$src1, VR128:$src2),
4670 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4671 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4672 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4673 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4674 (ins VR128:$src1, i128mem:$src2),
4676 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4677 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4679 (IntId128 VR128:$src1,
4680 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4683 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4684 let isCommutable = 0 in
4685 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4687 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4689 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4691 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4693 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4695 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4697 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4699 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4701 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4703 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4705 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4709 let Constraints = "$src1 = $dst" in {
4710 let isCommutable = 0 in
4711 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4712 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4713 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4714 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4715 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4716 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4717 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4718 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4719 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4720 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4721 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4724 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4725 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4726 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4727 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4729 /// SS48I_binop_rm - Simple SSE41 binary operator.
4730 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4731 ValueType OpVT, bit Is2Addr = 1> {
4732 let isCommutable = 1 in
4733 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4734 (ins VR128:$src1, VR128:$src2),
4736 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4737 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4738 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4740 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4741 (ins VR128:$src1, i128mem:$src2),
4743 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4744 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4745 [(set VR128:$dst, (OpNode VR128:$src1,
4746 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4750 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4751 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4752 let Constraints = "$src1 = $dst" in
4753 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4755 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4756 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4757 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4758 X86MemOperand x86memop, bit Is2Addr = 1> {
4759 let isCommutable = 1 in
4760 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4761 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4763 !strconcat(OpcodeStr,
4764 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4765 !strconcat(OpcodeStr,
4766 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4767 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4769 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4770 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4772 !strconcat(OpcodeStr,
4773 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4774 !strconcat(OpcodeStr,
4775 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4778 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4782 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4783 let isCommutable = 0 in {
4784 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4785 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4786 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4787 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4788 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4789 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4790 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4791 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4792 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4793 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4794 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4795 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4797 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4798 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4799 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4800 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4801 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4802 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4805 let Constraints = "$src1 = $dst" in {
4806 let isCommutable = 0 in {
4807 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4808 VR128, memopv16i8, i128mem>;
4809 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4810 VR128, memopv16i8, i128mem>;
4811 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4812 VR128, memopv16i8, i128mem>;
4813 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4814 VR128, memopv16i8, i128mem>;
4816 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4817 VR128, memopv16i8, i128mem>;
4818 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4819 VR128, memopv16i8, i128mem>;
4822 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4823 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4824 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4825 RegisterClass RC, X86MemOperand x86memop,
4826 PatFrag mem_frag, Intrinsic IntId> {
4827 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4828 (ins RC:$src1, RC:$src2, RC:$src3),
4829 !strconcat(OpcodeStr,
4830 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4831 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4832 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4834 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4835 (ins RC:$src1, x86memop:$src2, RC:$src3),
4836 !strconcat(OpcodeStr,
4837 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4839 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4841 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4845 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4846 memopv16i8, int_x86_sse41_blendvpd>;
4847 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4848 memopv16i8, int_x86_sse41_blendvps>;
4849 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4850 memopv16i8, int_x86_sse41_pblendvb>;
4851 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4852 memopv32i8, int_x86_avx_blendv_pd_256>;
4853 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4854 memopv32i8, int_x86_avx_blendv_ps_256>;
4856 /// SS41I_ternary_int - SSE 4.1 ternary operator
4857 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4858 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4859 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4860 (ins VR128:$src1, VR128:$src2),
4861 !strconcat(OpcodeStr,
4862 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4863 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4866 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4867 (ins VR128:$src1, i128mem:$src2),
4868 !strconcat(OpcodeStr,
4869 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4872 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4876 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4877 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4878 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4880 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4881 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4882 "vmovntdqa\t{$src, $dst|$dst, $src}",
4883 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4885 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4886 "movntdqa\t{$src, $dst|$dst, $src}",
4887 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4890 //===----------------------------------------------------------------------===//
4891 // SSE4.2 - Compare Instructions
4892 //===----------------------------------------------------------------------===//
4894 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4895 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4896 Intrinsic IntId128, bit Is2Addr = 1> {
4897 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4898 (ins VR128:$src1, VR128:$src2),
4900 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4901 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4902 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4904 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4905 (ins VR128:$src1, i128mem:$src2),
4907 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4908 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4910 (IntId128 VR128:$src1,
4911 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4914 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4915 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4917 let Constraints = "$src1 = $dst" in
4918 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4920 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4921 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4922 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4923 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4925 //===----------------------------------------------------------------------===//
4926 // SSE4.2 - String/text Processing Instructions
4927 //===----------------------------------------------------------------------===//
4929 // Packed Compare Implicit Length Strings, Return Mask
4930 multiclass pseudo_pcmpistrm<string asm> {
4931 def REG : PseudoI<(outs VR128:$dst),
4932 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4933 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4935 def MEM : PseudoI<(outs VR128:$dst),
4936 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4937 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4938 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4941 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4942 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4943 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4946 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4947 Predicates = [HasAVX] in {
4948 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4949 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4950 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4951 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4952 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4953 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4956 let Defs = [XMM0, EFLAGS] in {
4957 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4958 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4959 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4960 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4961 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4962 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4965 // Packed Compare Explicit Length Strings, Return Mask
4966 multiclass pseudo_pcmpestrm<string asm> {
4967 def REG : PseudoI<(outs VR128:$dst),
4968 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4969 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4970 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4971 def MEM : PseudoI<(outs VR128:$dst),
4972 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4973 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4974 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4977 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4978 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4979 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4982 let isAsmParserOnly = 1, Predicates = [HasAVX],
4983 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4984 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4985 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4986 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4987 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4988 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4989 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4992 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4993 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4994 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4995 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4996 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4997 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4998 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5001 // Packed Compare Implicit Length Strings, Return Index
5002 let Defs = [ECX, EFLAGS] in {
5003 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5004 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5005 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5006 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5007 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5008 (implicit EFLAGS)]>, OpSize;
5009 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5010 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5011 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5012 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5013 (implicit EFLAGS)]>, OpSize;
5017 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5018 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5020 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5022 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5024 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5026 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5028 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5032 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5033 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5034 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5035 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5036 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5037 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5039 // Packed Compare Explicit Length Strings, Return Index
5040 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5041 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5042 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5043 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5044 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5045 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5046 (implicit EFLAGS)]>, OpSize;
5047 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5048 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5049 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5051 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5052 (implicit EFLAGS)]>, OpSize;
5056 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5057 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5059 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5061 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5063 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5065 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5067 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5071 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5072 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5073 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5074 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5075 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5076 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5078 //===----------------------------------------------------------------------===//
5079 // SSE4.2 - CRC Instructions
5080 //===----------------------------------------------------------------------===//
5082 // No CRC instructions have AVX equivalents
5084 // crc intrinsic instruction
5085 // This set of instructions are only rm, the only difference is the size
5087 let Constraints = "$src1 = $dst" in {
5088 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5089 (ins GR32:$src1, i8mem:$src2),
5090 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5092 (int_x86_sse42_crc32_8 GR32:$src1,
5093 (load addr:$src2)))]>;
5094 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5095 (ins GR32:$src1, GR8:$src2),
5096 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5098 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5099 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5100 (ins GR32:$src1, i16mem:$src2),
5101 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5103 (int_x86_sse42_crc32_16 GR32:$src1,
5104 (load addr:$src2)))]>,
5106 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5107 (ins GR32:$src1, GR16:$src2),
5108 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5110 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5112 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5113 (ins GR32:$src1, i32mem:$src2),
5114 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5116 (int_x86_sse42_crc32_32 GR32:$src1,
5117 (load addr:$src2)))]>;
5118 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5119 (ins GR32:$src1, GR32:$src2),
5120 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5122 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5123 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5124 (ins GR64:$src1, i8mem:$src2),
5125 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5127 (int_x86_sse42_crc64_8 GR64:$src1,
5128 (load addr:$src2)))]>,
5130 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5131 (ins GR64:$src1, GR8:$src2),
5132 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5134 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5136 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5137 (ins GR64:$src1, i64mem:$src2),
5138 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5140 (int_x86_sse42_crc64_64 GR64:$src1,
5141 (load addr:$src2)))]>,
5143 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5144 (ins GR64:$src1, GR64:$src2),
5145 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5147 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5151 //===----------------------------------------------------------------------===//
5152 // AES-NI Instructions
5153 //===----------------------------------------------------------------------===//
5155 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5156 Intrinsic IntId128, bit Is2Addr = 1> {
5157 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5158 (ins VR128:$src1, VR128:$src2),
5160 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5161 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5162 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5164 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5165 (ins VR128:$src1, i128mem:$src2),
5167 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5168 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5170 (IntId128 VR128:$src1,
5171 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5174 // Perform One Round of an AES Encryption/Decryption Flow
5175 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5176 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5177 int_x86_aesni_aesenc, 0>, VEX_4V;
5178 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5179 int_x86_aesni_aesenclast, 0>, VEX_4V;
5180 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5181 int_x86_aesni_aesdec, 0>, VEX_4V;
5182 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5183 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5186 let Constraints = "$src1 = $dst" in {
5187 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5188 int_x86_aesni_aesenc>;
5189 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5190 int_x86_aesni_aesenclast>;
5191 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5192 int_x86_aesni_aesdec>;
5193 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5194 int_x86_aesni_aesdeclast>;
5197 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5198 (AESENCrr VR128:$src1, VR128:$src2)>;
5199 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5200 (AESENCrm VR128:$src1, addr:$src2)>;
5201 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5202 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5203 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5204 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5205 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5206 (AESDECrr VR128:$src1, VR128:$src2)>;
5207 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5208 (AESDECrm VR128:$src1, addr:$src2)>;
5209 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5210 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5211 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5212 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5214 // Perform the AES InvMixColumn Transformation
5215 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5216 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5218 "vaesimc\t{$src1, $dst|$dst, $src1}",
5220 (int_x86_aesni_aesimc VR128:$src1))]>,
5222 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5223 (ins i128mem:$src1),
5224 "vaesimc\t{$src1, $dst|$dst, $src1}",
5226 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5229 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5231 "aesimc\t{$src1, $dst|$dst, $src1}",
5233 (int_x86_aesni_aesimc VR128:$src1))]>,
5235 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5236 (ins i128mem:$src1),
5237 "aesimc\t{$src1, $dst|$dst, $src1}",
5239 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5242 // AES Round Key Generation Assist
5243 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5244 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5245 (ins VR128:$src1, i8imm:$src2),
5246 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5248 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5250 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5251 (ins i128mem:$src1, i8imm:$src2),
5252 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5254 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5258 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5259 (ins VR128:$src1, i8imm:$src2),
5260 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5262 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5264 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5265 (ins i128mem:$src1, i8imm:$src2),
5266 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5268 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5272 //===----------------------------------------------------------------------===//
5273 // CLMUL Instructions
5274 //===----------------------------------------------------------------------===//
5276 // Only the AVX version of CLMUL instructions are described here.
5278 // Carry-less Multiplication instructions
5279 let isAsmParserOnly = 1 in {
5280 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5281 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5282 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5285 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5286 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5287 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5291 multiclass avx_vpclmul<string asm> {
5292 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5293 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5296 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5297 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5300 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5301 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5302 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5303 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5305 } // isAsmParserOnly
5307 //===----------------------------------------------------------------------===//
5309 //===----------------------------------------------------------------------===//
5311 let isAsmParserOnly = 1 in {
5313 // Load from memory and broadcast to all elements of the destination operand
5314 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5315 X86MemOperand x86memop, Intrinsic Int> :
5316 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5317 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5318 [(set RC:$dst, (Int addr:$src))]>, VEX;
5320 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5321 int_x86_avx_vbroadcastss>;
5322 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5323 int_x86_avx_vbroadcastss_256>;
5324 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5325 int_x86_avx_vbroadcast_sd_256>;
5326 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5327 int_x86_avx_vbroadcastf128_pd_256>;
5329 // Insert packed floating-point values
5330 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5331 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5332 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5334 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5335 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5336 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5339 // Extract packed floating-point values
5340 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5341 (ins VR256:$src1, i8imm:$src2),
5342 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5344 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5345 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5346 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5349 // Conditional SIMD Packed Loads and Stores
5350 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5351 Intrinsic IntLd, Intrinsic IntLd256,
5352 Intrinsic IntSt, Intrinsic IntSt256,
5353 PatFrag pf128, PatFrag pf256> {
5354 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5355 (ins VR128:$src1, f128mem:$src2),
5356 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5357 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5359 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5360 (ins VR256:$src1, f256mem:$src2),
5361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5362 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5364 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5365 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5367 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5368 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5369 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5370 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5371 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5374 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5375 int_x86_avx_maskload_ps,
5376 int_x86_avx_maskload_ps_256,
5377 int_x86_avx_maskstore_ps,
5378 int_x86_avx_maskstore_ps_256,
5379 memopv4f32, memopv8f32>;
5380 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5381 int_x86_avx_maskload_pd,
5382 int_x86_avx_maskload_pd_256,
5383 int_x86_avx_maskstore_pd,
5384 int_x86_avx_maskstore_pd_256,
5385 memopv2f64, memopv4f64>;
5387 // Permute Floating-Point Values
5388 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5389 RegisterClass RC, X86MemOperand x86memop_f,
5390 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5391 Intrinsic IntVar, Intrinsic IntImm> {
5392 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5393 (ins RC:$src1, RC:$src2),
5394 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5395 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5396 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5397 (ins RC:$src1, x86memop_i:$src2),
5398 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5399 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5401 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5402 (ins RC:$src1, i8imm:$src2),
5403 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5404 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5405 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5406 (ins x86memop_f:$src1, i8imm:$src2),
5407 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5408 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5411 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5412 memopv4f32, memopv4i32,
5413 int_x86_avx_vpermilvar_ps,
5414 int_x86_avx_vpermil_ps>;
5415 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5416 memopv8f32, memopv8i32,
5417 int_x86_avx_vpermilvar_ps_256,
5418 int_x86_avx_vpermil_ps_256>;
5419 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5420 memopv2f64, memopv2i64,
5421 int_x86_avx_vpermilvar_pd,
5422 int_x86_avx_vpermil_pd>;
5423 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5424 memopv4f64, memopv4i64,
5425 int_x86_avx_vpermilvar_pd_256,
5426 int_x86_avx_vpermil_pd_256>;
5428 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5429 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5430 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5432 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5433 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5434 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5437 // Zero All YMM registers
5438 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5439 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5441 // Zero Upper bits of YMM registers
5442 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5443 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5445 } // isAsmParserOnly
5447 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5448 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5449 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5450 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5451 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5452 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5454 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5455 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5456 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5457 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5458 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5459 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5461 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5462 (VBROADCASTF128 addr:$src)>;
5464 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5465 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5466 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5467 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5468 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5469 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5471 def : Pat<(int_x86_avx_vperm2f128_ps_256
5472 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5473 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5474 def : Pat<(int_x86_avx_vperm2f128_pd_256
5475 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5476 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5477 def : Pat<(int_x86_avx_vperm2f128_si_256
5478 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5479 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5481 //===----------------------------------------------------------------------===//
5482 // SSE Shuffle pattern fragments
5483 //===----------------------------------------------------------------------===//
5485 // This is part of a "work in progress" refactoring. The idea is that all
5486 // vector shuffles are going to be translated into target specific nodes and
5487 // directly matched by the patterns below (which can be changed along the way)
5488 // The AVX version of some but not all of them are described here, and more
5489 // should come in a near future.
5491 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5492 // SSE2 loads, which are always promoted to v2i64. The last one should match
5493 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5494 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5495 // we investigate further.
5496 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5498 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5499 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5501 (PSHUFDmi addr:$src1, imm:$imm)>;
5502 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5504 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5506 // Shuffle with PSHUFD instruction.
5507 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5508 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5509 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5510 (PSHUFDri VR128:$src1, imm:$imm)>;
5512 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5513 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5514 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5515 (PSHUFDri VR128:$src1, imm:$imm)>;
5517 // Shuffle with SHUFPD instruction.
5518 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5519 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5520 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5521 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5522 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5523 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5525 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5526 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5527 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5528 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5530 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5531 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5532 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5533 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5535 // Shuffle with SHUFPS instruction.
5536 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5537 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5538 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5539 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5540 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5541 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5543 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5544 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5545 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5546 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5548 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5549 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5550 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5551 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5552 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5553 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5555 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5556 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5557 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5558 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5560 // Shuffle with MOVHLPS instruction
5561 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5562 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5563 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5564 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5566 // Shuffle with MOVDDUP instruction
5567 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5568 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5569 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5570 (MOVDDUPrm addr:$src)>;
5572 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5573 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5574 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5575 (MOVDDUPrm addr:$src)>;
5577 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5578 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5579 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5580 (MOVDDUPrm addr:$src)>;
5582 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5583 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5584 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5585 (MOVDDUPrm addr:$src)>;
5587 def : Pat<(X86Movddup (bc_v2f64
5588 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5589 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5590 def : Pat<(X86Movddup (bc_v2f64
5591 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5592 (MOVDDUPrm addr:$src)>;
5595 // Shuffle with UNPCKLPS
5596 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5597 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5598 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5599 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5601 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5602 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5603 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5604 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5606 // Shuffle with UNPCKHPS
5607 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5608 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5609 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5610 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5612 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5613 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5614 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5615 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5617 // Shuffle with UNPCKLPD
5618 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5619 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5620 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5621 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5623 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5624 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5625 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5626 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5628 // Shuffle with UNPCKHPD
5629 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5630 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5631 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5632 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5634 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5635 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5636 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5637 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5639 // Shuffle with PUNPCKLBW
5640 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5641 (bc_v16i8 (memopv2i64 addr:$src2)))),
5642 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5643 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5644 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5646 // Shuffle with PUNPCKLWD
5647 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5648 (bc_v8i16 (memopv2i64 addr:$src2)))),
5649 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5650 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5651 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5653 // Shuffle with PUNPCKLDQ
5654 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5655 (bc_v4i32 (memopv2i64 addr:$src2)))),
5656 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5657 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5658 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5660 // Shuffle with PUNPCKLQDQ
5661 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5662 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5663 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5664 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5666 // Shuffle with PUNPCKHBW
5667 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5668 (bc_v16i8 (memopv2i64 addr:$src2)))),
5669 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5670 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5671 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5673 // Shuffle with PUNPCKHWD
5674 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5675 (bc_v8i16 (memopv2i64 addr:$src2)))),
5676 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5677 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5678 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5680 // Shuffle with PUNPCKHDQ
5681 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5682 (bc_v4i32 (memopv2i64 addr:$src2)))),
5683 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5684 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5685 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5687 // Shuffle with PUNPCKHQDQ
5688 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5689 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5690 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5691 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5693 // Shuffle with MOVLHPS
5694 def : Pat<(X86Movlhps VR128:$src1,
5695 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5696 (MOVHPSrm VR128:$src1, addr:$src2)>;
5697 def : Pat<(X86Movlhps VR128:$src1,
5698 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5699 (MOVHPSrm VR128:$src1, addr:$src2)>;
5700 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5701 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5702 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5703 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5704 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5705 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5707 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5708 // is during lowering, where it's not possible to recognize the load fold cause
5709 // it has two uses through a bitcast. One use disappears at isel time and the
5710 // fold opportunity reappears.
5711 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5712 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5714 // Shuffle with MOVLHPD
5715 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5716 (scalar_to_vector (loadf64 addr:$src2)))),
5717 (MOVHPDrm VR128:$src1, addr:$src2)>;
5719 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5720 // is during lowering, where it's not possible to recognize the load fold cause
5721 // it has two uses through a bitcast. One use disappears at isel time and the
5722 // fold opportunity reappears.
5723 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5724 (scalar_to_vector (loadf64 addr:$src2)))),
5725 (MOVHPDrm VR128:$src1, addr:$src2)>;
5727 // Shuffle with MOVSS
5728 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5729 (MOVSSrr VR128:$src1, FR32:$src2)>;
5730 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5731 (MOVSSrr (v4i32 VR128:$src1),
5732 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5733 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5734 (MOVSSrr (v4f32 VR128:$src1),
5735 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5736 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5737 // is during lowering, where it's not possible to recognize the load fold cause
5738 // it has two uses through a bitcast. One use disappears at isel time and the
5739 // fold opportunity reappears.
5740 def : Pat<(X86Movss VR128:$src1,
5741 (bc_v4i32 (v2i64 (load addr:$src2)))),
5742 (MOVLPSrm VR128:$src1, addr:$src2)>;
5744 // Shuffle with MOVSD
5745 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5746 (MOVSDrr VR128:$src1, FR64:$src2)>;
5747 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5748 (MOVSDrr (v2i64 VR128:$src1),
5749 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5750 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5751 (MOVSDrr (v2f64 VR128:$src1),
5752 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5753 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5754 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5755 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5756 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5758 // Shuffle with MOVSHDUP
5759 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5760 (MOVSHDUPrr VR128:$src)>;
5761 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5762 (MOVSHDUPrm addr:$src)>;
5764 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5765 (MOVSHDUPrr VR128:$src)>;
5766 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5767 (MOVSHDUPrm addr:$src)>;
5769 // Shuffle with MOVSLDUP
5770 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5771 (MOVSLDUPrr VR128:$src)>;
5772 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5773 (MOVSLDUPrm addr:$src)>;
5775 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5776 (MOVSLDUPrr VR128:$src)>;
5777 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5778 (MOVSLDUPrm addr:$src)>;
5780 // Shuffle with PSHUFHW
5781 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5782 (PSHUFHWri VR128:$src, imm:$imm)>;
5783 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5784 (PSHUFHWmi addr:$src, imm:$imm)>;
5786 // Shuffle with PSHUFLW
5787 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5788 (PSHUFLWri VR128:$src, imm:$imm)>;
5789 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5790 (PSHUFLWmi addr:$src, imm:$imm)>;
5792 // Shuffle with PALIGN
5793 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5794 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5795 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5796 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5797 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5798 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5799 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5800 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5802 // Shuffle with MOVLPS
5803 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5804 (MOVLPSrm VR128:$src1, addr:$src2)>;
5805 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5806 (MOVLPSrm VR128:$src1, addr:$src2)>;
5807 def : Pat<(X86Movlps VR128:$src1,
5808 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5809 (MOVLPSrm VR128:$src1, addr:$src2)>;
5810 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5811 // is during lowering, where it's not possible to recognize the load fold cause
5812 // it has two uses through a bitcast. One use disappears at isel time and the
5813 // fold opportunity reappears.
5814 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5815 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5817 // Shuffle with MOVLPD
5818 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5819 (MOVLPDrm VR128:$src1, addr:$src2)>;
5820 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5821 (MOVLPDrm VR128:$src1, addr:$src2)>;
5822 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5823 (scalar_to_vector (loadf64 addr:$src2)))),
5824 (MOVLPDrm VR128:$src1, addr:$src2)>;
5826 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5827 def : Pat<(store (f64 (vector_extract
5828 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5829 (MOVHPSmr addr:$dst, VR128:$src)>;
5830 def : Pat<(store (f64 (vector_extract
5831 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5832 (MOVHPDmr addr:$dst, VR128:$src)>;
5834 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5835 (MOVLPSmr addr:$src1, VR128:$src2)>;
5836 def : Pat<(store (v4i32 (X86Movlps
5837 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5838 (MOVLPSmr addr:$src1, VR128:$src2)>;
5840 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5841 (MOVLPDmr addr:$src1, VR128:$src2)>;
5842 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5843 (MOVLPDmr addr:$src1, VR128:$src2)>;