1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 let isCommutable = 1 in
85 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
87 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
88 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
90 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
92 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
93 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
97 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
98 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
99 string asm, string SSEVer, string FPSizeStr,
100 X86MemOperand x86memop, PatFrag mem_frag,
101 Domain d, bit Is2Addr = 1> {
102 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
104 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
105 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
106 [(set RC:$dst, (!cast<Intrinsic>(
107 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
108 RC:$src1, RC:$src2))], d>;
109 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
111 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
112 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
113 [(set RC:$dst, (!cast<Intrinsic>(
114 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
115 RC:$src1, (mem_frag addr:$src2)))], d>;
118 //===----------------------------------------------------------------------===//
119 // SSE 1 & 2 - Move Instructions
120 //===----------------------------------------------------------------------===//
122 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
123 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
124 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
126 // Loading from memory automatically zeroing upper bits.
127 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
128 PatFrag mem_pat, string OpcodeStr> :
129 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
131 [(set RC:$dst, (mem_pat addr:$src))]>;
133 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
134 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
135 // is used instead. Register-to-register movss/movsd is not modeled as an
136 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
137 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
138 let isAsmParserOnly = 1 in {
139 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
140 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
141 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
142 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
144 let canFoldAsLoad = 1, isReMaterializable = 1 in {
145 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
147 let AddedComplexity = 20 in
148 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
152 let Constraints = "$src1 = $dst" in {
153 def MOVSSrr : sse12_move_rr<FR32, v4f32,
154 "movss\t{$src2, $dst|$dst, $src2}">, XS;
155 def MOVSDrr : sse12_move_rr<FR64, v2f64,
156 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
159 let canFoldAsLoad = 1, isReMaterializable = 1 in {
160 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
162 let AddedComplexity = 20 in
163 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
166 let AddedComplexity = 15 in {
167 // Extract the low 32-bit value from one vector and insert it into another.
168 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
169 (MOVSSrr (v4f32 VR128:$src1),
170 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
171 // Extract the low 64-bit value from one vector and insert it into another.
172 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
173 (MOVSDrr (v2f64 VR128:$src1),
174 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
177 // Implicitly promote a 32-bit scalar to a vector.
178 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
179 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
180 // Implicitly promote a 64-bit scalar to a vector.
181 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
182 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
183 // Implicitly promote a 32-bit scalar to a vector.
184 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
185 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
186 // Implicitly promote a 64-bit scalar to a vector.
187 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
188 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
190 let AddedComplexity = 20 in {
191 // MOVSSrm zeros the high parts of the register; represent this
192 // with SUBREG_TO_REG.
193 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
194 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
195 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
196 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
197 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
198 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
199 // MOVSDrm zeros the high parts of the register; represent this
200 // with SUBREG_TO_REG.
201 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
202 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
203 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
204 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
205 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
206 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
207 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
208 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
209 def : Pat<(v2f64 (X86vzload addr:$src)),
210 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
213 // Store scalar value to memory.
214 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
215 "movss\t{$src, $dst|$dst, $src}",
216 [(store FR32:$src, addr:$dst)]>;
217 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
218 "movsd\t{$src, $dst|$dst, $src}",
219 [(store FR64:$src, addr:$dst)]>;
221 let isAsmParserOnly = 1 in {
222 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
223 "movss\t{$src, $dst|$dst, $src}",
224 [(store FR32:$src, addr:$dst)]>, XS, VEX;
225 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
226 "movsd\t{$src, $dst|$dst, $src}",
227 [(store FR64:$src, addr:$dst)]>, XD, VEX;
230 // Extract and store.
231 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
234 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
235 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
238 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
240 // Move Aligned/Unaligned floating point values
241 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
242 X86MemOperand x86memop, PatFrag ld_frag,
243 string asm, Domain d,
244 bit IsReMaterializable = 1> {
245 let neverHasSideEffects = 1 in
246 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
247 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
248 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
249 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
250 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
251 [(set RC:$dst, (ld_frag addr:$src))], d>;
254 let isAsmParserOnly = 1 in {
255 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
256 "movaps", SSEPackedSingle>, VEX;
257 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
258 "movapd", SSEPackedDouble>, OpSize, VEX;
259 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
260 "movups", SSEPackedSingle>, VEX;
261 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
262 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
264 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
265 "movaps", SSEPackedSingle>, VEX;
266 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
267 "movapd", SSEPackedDouble>, OpSize, VEX;
268 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
269 "movups", SSEPackedSingle>, VEX;
270 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
271 "movupd", SSEPackedDouble, 0>, OpSize, VEX;
273 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
274 "movaps", SSEPackedSingle>, TB;
275 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
276 "movapd", SSEPackedDouble>, TB, OpSize;
277 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
278 "movups", SSEPackedSingle>, TB;
279 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
280 "movupd", SSEPackedDouble, 0>, TB, OpSize;
282 let isAsmParserOnly = 1 in {
283 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
284 "movaps\t{$src, $dst|$dst, $src}",
285 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
286 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
287 "movapd\t{$src, $dst|$dst, $src}",
288 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
289 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
290 "movups\t{$src, $dst|$dst, $src}",
291 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
292 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
293 "movupd\t{$src, $dst|$dst, $src}",
294 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
295 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
296 "movaps\t{$src, $dst|$dst, $src}",
297 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
298 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
299 "movapd\t{$src, $dst|$dst, $src}",
300 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
301 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
302 "movups\t{$src, $dst|$dst, $src}",
303 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
304 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
305 "movupd\t{$src, $dst|$dst, $src}",
306 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
309 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
310 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
311 (VMOVUPSYmr addr:$dst, VR256:$src)>;
313 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
314 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
315 (VMOVUPDYmr addr:$dst, VR256:$src)>;
317 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
318 "movaps\t{$src, $dst|$dst, $src}",
319 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
320 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
321 "movapd\t{$src, $dst|$dst, $src}",
322 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
323 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
324 "movups\t{$src, $dst|$dst, $src}",
325 [(store (v4f32 VR128:$src), addr:$dst)]>;
326 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
327 "movupd\t{$src, $dst|$dst, $src}",
328 [(store (v2f64 VR128:$src), addr:$dst)]>;
330 // Intrinsic forms of MOVUPS/D load and store
331 let isAsmParserOnly = 1 in {
332 let canFoldAsLoad = 1, isReMaterializable = 1 in
333 def VMOVUPSrm_Int : VPSI<0x10, MRMSrcMem, (outs VR128:$dst),
335 "movups\t{$src, $dst|$dst, $src}",
336 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>, VEX;
337 def VMOVUPDrm_Int : VPDI<0x10, MRMSrcMem, (outs VR128:$dst),
339 "movupd\t{$src, $dst|$dst, $src}",
340 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>, VEX;
341 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
342 (ins f128mem:$dst, VR128:$src),
343 "movups\t{$src, $dst|$dst, $src}",
344 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
345 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
346 (ins f128mem:$dst, VR128:$src),
347 "movupd\t{$src, $dst|$dst, $src}",
348 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
350 let canFoldAsLoad = 1, isReMaterializable = 1 in
351 def MOVUPSrm_Int : PSI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
352 "movups\t{$src, $dst|$dst, $src}",
353 [(set VR128:$dst, (int_x86_sse_loadu_ps addr:$src))]>;
354 def MOVUPDrm_Int : PDI<0x10, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
355 "movupd\t{$src, $dst|$dst, $src}",
356 [(set VR128:$dst, (int_x86_sse2_loadu_pd addr:$src))]>;
358 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
359 "movups\t{$src, $dst|$dst, $src}",
360 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
361 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
362 "movupd\t{$src, $dst|$dst, $src}",
363 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
365 // Move Low/High packed floating point values
366 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
367 PatFrag mov_frag, string base_opc,
369 def PSrm : PI<opc, MRMSrcMem,
370 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
371 !strconcat(base_opc, "s", asm_opr),
374 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
375 SSEPackedSingle>, TB;
377 def PDrm : PI<opc, MRMSrcMem,
378 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
379 !strconcat(base_opc, "d", asm_opr),
380 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
381 (scalar_to_vector (loadf64 addr:$src2)))))],
382 SSEPackedDouble>, TB, OpSize;
385 let isAsmParserOnly = 1, AddedComplexity = 20 in {
386 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
387 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
388 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
389 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
391 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
392 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
393 "\t{$src2, $dst|$dst, $src2}">;
394 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
395 "\t{$src2, $dst|$dst, $src2}">;
398 let isAsmParserOnly = 1 in {
399 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
400 "movlps\t{$src, $dst|$dst, $src}",
401 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
402 (iPTR 0))), addr:$dst)]>, VEX;
403 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
404 "movlpd\t{$src, $dst|$dst, $src}",
405 [(store (f64 (vector_extract (v2f64 VR128:$src),
406 (iPTR 0))), addr:$dst)]>, VEX;
408 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
409 "movlps\t{$src, $dst|$dst, $src}",
410 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
411 (iPTR 0))), addr:$dst)]>;
412 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
413 "movlpd\t{$src, $dst|$dst, $src}",
414 [(store (f64 (vector_extract (v2f64 VR128:$src),
415 (iPTR 0))), addr:$dst)]>;
417 // v2f64 extract element 1 is always custom lowered to unpack high to low
418 // and extract element 0 so the non-store version isn't too horrible.
419 let isAsmParserOnly = 1 in {
420 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
421 "movhps\t{$src, $dst|$dst, $src}",
422 [(store (f64 (vector_extract
423 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
424 (undef)), (iPTR 0))), addr:$dst)]>,
426 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
427 "movhpd\t{$src, $dst|$dst, $src}",
428 [(store (f64 (vector_extract
429 (v2f64 (unpckh VR128:$src, (undef))),
430 (iPTR 0))), addr:$dst)]>,
433 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
434 "movhps\t{$src, $dst|$dst, $src}",
435 [(store (f64 (vector_extract
436 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
437 (undef)), (iPTR 0))), addr:$dst)]>;
438 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
439 "movhpd\t{$src, $dst|$dst, $src}",
440 [(store (f64 (vector_extract
441 (v2f64 (unpckh VR128:$src, (undef))),
442 (iPTR 0))), addr:$dst)]>;
444 let isAsmParserOnly = 1, AddedComplexity = 20 in {
445 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
446 (ins VR128:$src1, VR128:$src2),
447 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
449 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
451 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
452 (ins VR128:$src1, VR128:$src2),
453 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
455 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
458 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
459 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
460 (ins VR128:$src1, VR128:$src2),
461 "movlhps\t{$src2, $dst|$dst, $src2}",
463 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
464 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
465 (ins VR128:$src1, VR128:$src2),
466 "movhlps\t{$src2, $dst|$dst, $src2}",
468 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
471 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
472 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
473 let AddedComplexity = 20 in {
474 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
475 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
476 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
477 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
480 //===----------------------------------------------------------------------===//
481 // SSE 1 & 2 - Conversion Instructions
482 //===----------------------------------------------------------------------===//
484 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
485 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
487 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
488 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
489 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
490 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
493 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
494 X86MemOperand x86memop, string asm> {
495 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
497 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
501 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
502 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
503 string asm, Domain d> {
504 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
505 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
506 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
507 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
510 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
511 X86MemOperand x86memop, string asm> {
512 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
513 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
514 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
515 (ins DstRC:$src1, x86memop:$src),
516 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
519 let isAsmParserOnly = 1 in {
520 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
521 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
522 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
523 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
525 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
526 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
527 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
528 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
531 // The assembler can recognize rr 64-bit instructions by seeing a rxx
532 // register, but the same isn't true when only using memory operands,
533 // provide other assembly "l" and "q" forms to address this explicitly
534 // where appropriate to do so.
535 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
537 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
539 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
541 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
543 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
547 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
548 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
549 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
550 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
551 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
552 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
553 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
554 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
555 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
556 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
557 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
558 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
559 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
560 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
561 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
562 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
564 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
565 // and/or XMM operand(s).
567 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
568 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
570 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
571 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
572 [(set DstRC:$dst, (Int SrcRC:$src))]>;
573 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
574 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
575 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
578 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
579 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
580 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
581 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
583 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
584 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
585 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
586 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
587 (ins DstRC:$src1, x86memop:$src2),
589 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
590 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
591 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
594 let isAsmParserOnly = 1 in {
595 defm Int_VCVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
596 f32mem, load, "cvtss2si">, XS, VEX;
597 defm Int_VCVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
598 int_x86_sse_cvtss2si64, f32mem, load, "cvtss2si">,
600 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
601 f128mem, load, "cvtsd2si">, XD, VEX;
602 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
603 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
606 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
607 // Get rid of this hack or rename the intrinsics, there are several
608 // intructions that only match with the intrinsic form, why create duplicates
609 // to let them be recognized by the assembler?
610 defm VCVTSD2SI_alt : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
611 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
612 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
613 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
615 defm Int_CVTSS2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse_cvtss2si,
616 f32mem, load, "cvtss2si">, XS;
617 defm Int_CVTSS2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse_cvtss2si64,
618 f32mem, load, "cvtss2si{q}">, XS, REX_W;
619 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
620 f128mem, load, "cvtsd2si{l}">, XD;
621 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
622 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
625 let isAsmParserOnly = 1 in {
626 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
627 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
628 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
629 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
631 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
632 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
633 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
634 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
638 let Constraints = "$src1 = $dst" in {
639 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
640 int_x86_sse_cvtsi2ss, i32mem, loadi32,
642 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
643 int_x86_sse_cvtsi642ss, i64mem, loadi64,
644 "cvtsi2ss{q}">, XS, REX_W;
645 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
646 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
648 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
649 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
650 "cvtsi2sd">, XD, REX_W;
655 // Aliases for intrinsics
656 let isAsmParserOnly = 1 in {
657 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
658 f32mem, load, "cvttss2si">, XS, VEX;
659 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
660 int_x86_sse_cvttss2si64, f32mem, load,
661 "cvttss2si">, XS, VEX, VEX_W;
662 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
663 f128mem, load, "cvttsd2si">, XD, VEX;
664 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
665 int_x86_sse2_cvttsd2si64, f128mem, load,
666 "cvttsd2si">, XD, VEX, VEX_W;
668 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
669 f32mem, load, "cvttss2si">, XS;
670 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
671 int_x86_sse_cvttss2si64, f32mem, load,
672 "cvttss2si{q}">, XS, REX_W;
673 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
674 f128mem, load, "cvttsd2si">, XD;
675 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
676 int_x86_sse2_cvttsd2si64, f128mem, load,
677 "cvttsd2si{q}">, XD, REX_W;
679 let isAsmParserOnly = 1, Pattern = []<dag> in {
680 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
681 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
682 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
683 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
685 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
686 "cvtdq2ps\t{$src, $dst|$dst, $src}",
687 SSEPackedSingle>, TB, VEX;
688 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
689 "cvtdq2ps\t{$src, $dst|$dst, $src}",
690 SSEPackedSingle>, TB, VEX;
692 let Pattern = []<dag> in {
693 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
694 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
695 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
696 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
697 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
698 "cvtdq2ps\t{$src, $dst|$dst, $src}",
699 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
704 // Convert scalar double to scalar single
705 let isAsmParserOnly = 1 in {
706 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
707 (ins FR64:$src1, FR64:$src2),
708 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
710 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
711 (ins FR64:$src1, f64mem:$src2),
712 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
713 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
715 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
718 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
719 "cvtsd2ss\t{$src, $dst|$dst, $src}",
720 [(set FR32:$dst, (fround FR64:$src))]>;
721 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
722 "cvtsd2ss\t{$src, $dst|$dst, $src}",
723 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
724 Requires<[HasSSE2, OptForSize]>;
726 let isAsmParserOnly = 1 in
727 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
728 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
730 let Constraints = "$src1 = $dst" in
731 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
732 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
734 // Convert scalar single to scalar double
735 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
736 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
737 (ins FR32:$src1, FR32:$src2),
738 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
739 []>, XS, Requires<[HasAVX]>, VEX_4V;
740 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
741 (ins FR32:$src1, f32mem:$src2),
742 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
743 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
745 def : Pat<(f64 (fextend FR32:$src)), (VCVTSS2SDrr FR32:$src, FR32:$src)>,
748 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
749 "cvtss2sd\t{$src, $dst|$dst, $src}",
750 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
752 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
753 "cvtss2sd\t{$src, $dst|$dst, $src}",
754 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
755 Requires<[HasSSE2, OptForSize]>;
757 let isAsmParserOnly = 1 in {
758 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
759 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
760 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
761 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
762 VR128:$src2))]>, XS, VEX_4V,
764 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
765 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
766 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
767 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
768 (load addr:$src2)))]>, XS, VEX_4V,
771 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
772 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
773 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
774 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
775 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
778 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
779 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
780 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
781 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
782 (load addr:$src2)))]>, XS,
786 def : Pat<(extloadf32 addr:$src),
787 (CVTSS2SDrr (MOVSSrm addr:$src))>,
788 Requires<[HasSSE2, OptForSpeed]>;
790 // Convert doubleword to packed single/double fp
791 let isAsmParserOnly = 1 in { // SSE2 instructions without OpSize prefix
792 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
793 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
794 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
795 TB, VEX, Requires<[HasAVX]>;
796 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
797 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
798 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
799 (bitconvert (memopv2i64 addr:$src))))]>,
800 TB, VEX, Requires<[HasAVX]>;
802 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
803 "cvtdq2ps\t{$src, $dst|$dst, $src}",
804 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
805 TB, Requires<[HasSSE2]>;
806 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
807 "cvtdq2ps\t{$src, $dst|$dst, $src}",
808 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
809 (bitconvert (memopv2i64 addr:$src))))]>,
810 TB, Requires<[HasSSE2]>;
812 // FIXME: why the non-intrinsic version is described as SSE3?
813 let isAsmParserOnly = 1 in { // SSE2 instructions with XS prefix
814 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
815 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
816 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
817 XS, VEX, Requires<[HasAVX]>;
818 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
819 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
820 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
821 (bitconvert (memopv2i64 addr:$src))))]>,
822 XS, VEX, Requires<[HasAVX]>;
824 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
825 "cvtdq2pd\t{$src, $dst|$dst, $src}",
826 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
827 XS, Requires<[HasSSE2]>;
828 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
829 "cvtdq2pd\t{$src, $dst|$dst, $src}",
830 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
831 (bitconvert (memopv2i64 addr:$src))))]>,
832 XS, Requires<[HasSSE2]>;
835 // Convert packed single/double fp to doubleword
836 let isAsmParserOnly = 1 in {
837 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
838 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
839 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
840 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
841 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
842 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
843 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
844 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
846 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
847 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
848 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
849 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
851 let isAsmParserOnly = 1 in {
852 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
853 "cvtps2dq\t{$src, $dst|$dst, $src}",
854 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
856 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
858 "cvtps2dq\t{$src, $dst|$dst, $src}",
859 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
860 (memop addr:$src)))]>, VEX;
862 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
863 "cvtps2dq\t{$src, $dst|$dst, $src}",
864 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
865 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
866 "cvtps2dq\t{$src, $dst|$dst, $src}",
867 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
868 (memop addr:$src)))]>;
870 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XD prefix
871 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
872 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
873 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
874 XD, VEX, Requires<[HasAVX]>;
875 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
876 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
877 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
878 (memop addr:$src)))]>,
879 XD, VEX, Requires<[HasAVX]>;
881 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
882 "cvtpd2dq\t{$src, $dst|$dst, $src}",
883 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
884 XD, Requires<[HasSSE2]>;
885 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
886 "cvtpd2dq\t{$src, $dst|$dst, $src}",
887 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
888 (memop addr:$src)))]>,
889 XD, Requires<[HasSSE2]>;
892 // Convert with truncation packed single/double fp to doubleword
893 let isAsmParserOnly = 1 in { // SSE2 packed instructions with XS prefix
894 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
895 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
896 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
897 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
898 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
899 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
900 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
901 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
903 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
904 "cvttps2dq\t{$src, $dst|$dst, $src}",
906 (int_x86_sse2_cvttps2dq VR128:$src))]>;
907 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
908 "cvttps2dq\t{$src, $dst|$dst, $src}",
910 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
913 let isAsmParserOnly = 1 in {
914 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
915 "vcvttps2dq\t{$src, $dst|$dst, $src}",
917 (int_x86_sse2_cvttps2dq VR128:$src))]>,
918 XS, VEX, Requires<[HasAVX]>;
919 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
920 "vcvttps2dq\t{$src, $dst|$dst, $src}",
921 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
922 (memop addr:$src)))]>,
923 XS, VEX, Requires<[HasAVX]>;
926 let isAsmParserOnly = 1 in {
927 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
929 "cvttpd2dq\t{$src, $dst|$dst, $src}",
930 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
932 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
934 "cvttpd2dq\t{$src, $dst|$dst, $src}",
935 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
936 (memop addr:$src)))]>, VEX;
938 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
939 "cvttpd2dq\t{$src, $dst|$dst, $src}",
940 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
941 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
942 "cvttpd2dq\t{$src, $dst|$dst, $src}",
943 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
944 (memop addr:$src)))]>;
946 let isAsmParserOnly = 1 in {
947 // The assembler can recognize rr 256-bit instructions by seeing a ymm
948 // register, but the same isn't true when using memory operands instead.
949 // Provide other assembly rr and rm forms to address this explicitly.
950 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
951 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
952 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
953 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
956 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
957 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
958 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
959 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
962 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
963 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
964 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
965 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
968 // Convert packed single to packed double
969 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
970 // SSE2 instructions without OpSize prefix
971 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
972 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
973 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
974 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
975 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
976 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
977 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
978 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, VEX;
980 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
981 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
982 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
983 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
985 let isAsmParserOnly = 1 in {
986 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
987 "vcvtps2pd\t{$src, $dst|$dst, $src}",
988 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
989 VEX, Requires<[HasAVX]>;
990 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
991 "vcvtps2pd\t{$src, $dst|$dst, $src}",
992 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
993 (load addr:$src)))]>,
994 VEX, Requires<[HasAVX]>;
996 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
997 "cvtps2pd\t{$src, $dst|$dst, $src}",
998 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
999 TB, Requires<[HasSSE2]>;
1000 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1001 "cvtps2pd\t{$src, $dst|$dst, $src}",
1002 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1003 (load addr:$src)))]>,
1004 TB, Requires<[HasSSE2]>;
1006 // Convert packed double to packed single
1007 let isAsmParserOnly = 1 in {
1008 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1009 // register, but the same isn't true when using memory operands instead.
1010 // Provide other assembly rr and rm forms to address this explicitly.
1011 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1012 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1013 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1014 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1017 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1018 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1019 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1020 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1023 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1024 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1025 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1026 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1028 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1029 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1030 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1031 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1034 let isAsmParserOnly = 1 in {
1035 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1036 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1037 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1038 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1040 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1041 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1042 (memop addr:$src)))]>;
1044 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1045 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1046 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1047 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1048 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1049 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1050 (memop addr:$src)))]>;
1052 // AVX 256-bit register conversion intrinsics
1053 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1054 // whenever possible to avoid declaring two versions of each one.
1055 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1056 (VCVTDQ2PSYrr VR256:$src)>;
1057 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1058 (VCVTDQ2PSYrm addr:$src)>;
1060 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1061 (VCVTPD2PSYrr VR256:$src)>;
1062 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1063 (VCVTPD2PSYrm addr:$src)>;
1065 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1066 (VCVTPS2DQYrr VR256:$src)>;
1067 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1068 (VCVTPS2DQYrm addr:$src)>;
1070 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1071 (VCVTPS2PDYrr VR128:$src)>;
1072 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1073 (VCVTPS2PDYrm addr:$src)>;
1075 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1076 (VCVTTPD2DQYrr VR256:$src)>;
1077 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1078 (VCVTTPD2DQYrm addr:$src)>;
1080 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1081 (VCVTTPS2DQYrr VR256:$src)>;
1082 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1083 (VCVTTPS2DQYrm addr:$src)>;
1085 //===----------------------------------------------------------------------===//
1086 // SSE 1 & 2 - Compare Instructions
1087 //===----------------------------------------------------------------------===//
1089 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1090 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1091 string asm, string asm_alt> {
1092 def rr : SIi8<0xC2, MRMSrcReg,
1093 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1096 def rm : SIi8<0xC2, MRMSrcMem,
1097 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1099 // Accept explicit immediate argument form instead of comparison code.
1100 let isAsmParserOnly = 1 in {
1101 def rr_alt : SIi8<0xC2, MRMSrcReg,
1102 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1105 def rm_alt : SIi8<0xC2, MRMSrcMem,
1106 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1111 let neverHasSideEffects = 1, isAsmParserOnly = 1 in {
1112 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1113 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1114 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1116 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1117 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1118 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1122 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1123 defm CMPSS : sse12_cmp_scalar<FR32, f32mem,
1124 "cmp${cc}ss\t{$src, $dst|$dst, $src}",
1125 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}">, XS;
1126 defm CMPSD : sse12_cmp_scalar<FR64, f64mem,
1127 "cmp${cc}sd\t{$src, $dst|$dst, $src}",
1128 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}">, XD;
1131 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1132 Intrinsic Int, string asm> {
1133 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1134 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1135 [(set VR128:$dst, (Int VR128:$src1,
1136 VR128:$src, imm:$cc))]>;
1137 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1138 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1139 [(set VR128:$dst, (Int VR128:$src1,
1140 (load addr:$src), imm:$cc))]>;
1143 // Aliases to match intrinsics which expect XMM operand(s).
1144 let isAsmParserOnly = 1 in {
1145 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1146 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1148 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1149 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1152 let Constraints = "$src1 = $dst" in {
1153 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1154 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1155 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1156 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1160 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1161 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1162 ValueType vt, X86MemOperand x86memop,
1163 PatFrag ld_frag, string OpcodeStr, Domain d> {
1164 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1165 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1166 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1167 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1168 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1169 [(set EFLAGS, (OpNode (vt RC:$src1),
1170 (ld_frag addr:$src2)))], d>;
1173 let Defs = [EFLAGS] in {
1174 let isAsmParserOnly = 1 in {
1175 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1176 "ucomiss", SSEPackedSingle>, VEX;
1177 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1178 "ucomisd", SSEPackedDouble>, OpSize, VEX;
1179 let Pattern = []<dag> in {
1180 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1181 "comiss", SSEPackedSingle>, VEX;
1182 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1183 "comisd", SSEPackedDouble>, OpSize, VEX;
1186 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1187 load, "ucomiss", SSEPackedSingle>, VEX;
1188 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1189 load, "ucomisd", SSEPackedDouble>, OpSize, VEX;
1191 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1192 load, "comiss", SSEPackedSingle>, VEX;
1193 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1194 load, "comisd", SSEPackedDouble>, OpSize, VEX;
1196 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1197 "ucomiss", SSEPackedSingle>, TB;
1198 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1199 "ucomisd", SSEPackedDouble>, TB, OpSize;
1201 let Pattern = []<dag> in {
1202 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1203 "comiss", SSEPackedSingle>, TB;
1204 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1205 "comisd", SSEPackedDouble>, TB, OpSize;
1208 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1209 load, "ucomiss", SSEPackedSingle>, TB;
1210 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1211 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1213 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1214 "comiss", SSEPackedSingle>, TB;
1215 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1216 "comisd", SSEPackedDouble>, TB, OpSize;
1217 } // Defs = [EFLAGS]
1219 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1220 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1221 Intrinsic Int, string asm, string asm_alt,
1223 def rri : PIi8<0xC2, MRMSrcReg,
1224 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1225 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1226 def rmi : PIi8<0xC2, MRMSrcMem,
1227 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1228 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1229 // Accept explicit immediate argument form instead of comparison code.
1230 let isAsmParserOnly = 1 in {
1231 def rri_alt : PIi8<0xC2, MRMSrcReg,
1232 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1234 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1235 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1240 let isAsmParserOnly = 1 in {
1241 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1242 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1243 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1244 SSEPackedSingle>, VEX_4V;
1245 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1246 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1247 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1248 SSEPackedDouble>, OpSize, VEX_4V;
1249 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1250 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1251 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1252 SSEPackedSingle>, VEX_4V;
1253 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1254 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1255 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1256 SSEPackedDouble>, OpSize, VEX_4V;
1258 let Constraints = "$src1 = $dst" in {
1259 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1260 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1261 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1262 SSEPackedSingle>, TB;
1263 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1264 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1265 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1266 SSEPackedDouble>, TB, OpSize;
1269 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1270 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1271 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1272 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1273 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1274 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1275 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1276 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1278 //===----------------------------------------------------------------------===//
1279 // SSE 1 & 2 - Shuffle Instructions
1280 //===----------------------------------------------------------------------===//
1282 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1283 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1284 ValueType vt, string asm, PatFrag mem_frag,
1285 Domain d, bit IsConvertibleToThreeAddress = 0> {
1286 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1287 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1288 [(set RC:$dst, (vt (shufp:$src3
1289 RC:$src1, (mem_frag addr:$src2))))], d>;
1290 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1291 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1292 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1294 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1297 let isAsmParserOnly = 1 in {
1298 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1299 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1300 memopv4f32, SSEPackedSingle>, VEX_4V;
1301 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1302 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1303 memopv8f32, SSEPackedSingle>, VEX_4V;
1304 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1305 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1306 memopv2f64, SSEPackedDouble>, OpSize, VEX_4V;
1307 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1308 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1309 memopv4f64, SSEPackedDouble>, OpSize, VEX_4V;
1312 let Constraints = "$src1 = $dst" in {
1313 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1314 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1315 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1317 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1318 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1319 memopv2f64, SSEPackedDouble>, TB, OpSize;
1322 //===----------------------------------------------------------------------===//
1323 // SSE 1 & 2 - Unpack Instructions
1324 //===----------------------------------------------------------------------===//
1326 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1327 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1328 PatFrag mem_frag, RegisterClass RC,
1329 X86MemOperand x86memop, string asm,
1331 def rr : PI<opc, MRMSrcReg,
1332 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1334 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1335 def rm : PI<opc, MRMSrcMem,
1336 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1338 (vt (OpNode RC:$src1,
1339 (mem_frag addr:$src2))))], d>;
1342 let AddedComplexity = 10 in {
1343 let isAsmParserOnly = 1 in {
1344 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1345 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1346 SSEPackedSingle>, VEX_4V;
1347 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1348 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1349 SSEPackedDouble>, OpSize, VEX_4V;
1350 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1351 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1352 SSEPackedSingle>, VEX_4V;
1353 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1354 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1355 SSEPackedDouble>, OpSize, VEX_4V;
1357 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1358 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1359 SSEPackedSingle>, VEX_4V;
1360 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1361 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1362 SSEPackedDouble>, OpSize, VEX_4V;
1363 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1364 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1365 SSEPackedSingle>, VEX_4V;
1366 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1367 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1368 SSEPackedDouble>, OpSize, VEX_4V;
1371 let Constraints = "$src1 = $dst" in {
1372 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1373 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1374 SSEPackedSingle>, TB;
1375 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1376 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1377 SSEPackedDouble>, TB, OpSize;
1378 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1379 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1380 SSEPackedSingle>, TB;
1381 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1382 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1383 SSEPackedDouble>, TB, OpSize;
1384 } // Constraints = "$src1 = $dst"
1385 } // AddedComplexity
1387 //===----------------------------------------------------------------------===//
1388 // SSE 1 & 2 - Extract Floating-Point Sign mask
1389 //===----------------------------------------------------------------------===//
1391 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1392 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1394 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1395 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1396 [(set GR32:$dst, (Int RC:$src))], d>;
1397 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1398 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1402 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1403 SSEPackedSingle>, TB;
1404 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1405 SSEPackedDouble>, TB, OpSize;
1407 let isAsmParserOnly = 1 in {
1408 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1409 "movmskps", SSEPackedSingle>, VEX;
1410 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1411 "movmskpd", SSEPackedDouble>, OpSize,
1413 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1414 "movmskps", SSEPackedSingle>, VEX;
1415 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1416 "movmskpd", SSEPackedDouble>, OpSize,
1420 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1421 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1422 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1423 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1425 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1426 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, VEX;
1427 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1428 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, OpSize,
1432 //===----------------------------------------------------------------------===//
1433 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
1434 //===----------------------------------------------------------------------===//
1436 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
1437 // names that start with 'Fs'.
1439 // Alias instructions that map fld0 to pxor for sse.
1440 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
1441 canFoldAsLoad = 1 in {
1442 // FIXME: Set encoding to pseudo!
1443 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1444 [(set FR32:$dst, fp32imm0)]>,
1445 Requires<[HasSSE1]>, TB, OpSize;
1446 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1447 [(set FR64:$dst, fpimm0)]>,
1448 Requires<[HasSSE2]>, TB, OpSize;
1449 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
1450 [(set FR32:$dst, fp32imm0)]>,
1451 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1452 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
1453 [(set FR64:$dst, fpimm0)]>,
1454 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
1457 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
1458 // bits are disregarded.
1459 let neverHasSideEffects = 1 in {
1460 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1461 "movaps\t{$src, $dst|$dst, $src}", []>;
1462 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1463 "movapd\t{$src, $dst|$dst, $src}", []>;
1466 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
1467 // bits are disregarded.
1468 let canFoldAsLoad = 1, isReMaterializable = 1 in {
1469 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
1470 "movaps\t{$src, $dst|$dst, $src}",
1471 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
1472 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
1473 "movapd\t{$src, $dst|$dst, $src}",
1474 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
1477 //===----------------------------------------------------------------------===//
1478 // SSE 1 & 2 - Logical Instructions
1479 //===----------------------------------------------------------------------===//
1481 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
1483 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
1485 let isAsmParserOnly = 1 in {
1486 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
1487 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, VEX_4V;
1489 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
1490 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, OpSize, VEX_4V;
1493 let Constraints = "$src1 = $dst" in {
1494 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
1495 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
1497 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
1498 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
1502 // Alias bitwise logical operations using SSE logical ops on packed FP values.
1503 let mayLoad = 0 in {
1504 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
1505 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
1506 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
1509 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
1510 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
1512 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
1514 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
1515 SDNode OpNode, int HasPat = 0,
1516 list<list<dag>> Pattern = []> {
1517 let isAsmParserOnly = 1, Pattern = []<dag> in {
1518 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1519 !strconcat(OpcodeStr, "ps"), f128mem,
1520 !if(HasPat, Pattern[0], // rr
1521 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1523 !if(HasPat, Pattern[2], // rm
1524 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1525 (memopv2i64 addr:$src2)))]), 0>,
1528 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1529 !strconcat(OpcodeStr, "pd"), f128mem,
1530 !if(HasPat, Pattern[1], // rr
1531 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1534 !if(HasPat, Pattern[3], // rm
1535 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1536 (memopv2i64 addr:$src2)))]), 0>,
1539 let Constraints = "$src1 = $dst" in {
1540 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
1541 !strconcat(OpcodeStr, "ps"), f128mem,
1542 !if(HasPat, Pattern[0], // rr
1543 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1,
1545 !if(HasPat, Pattern[2], // rm
1546 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
1547 (memopv2i64 addr:$src2)))])>, TB;
1549 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
1550 !strconcat(OpcodeStr, "pd"), f128mem,
1551 !if(HasPat, Pattern[1], // rr
1552 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1555 !if(HasPat, Pattern[3], // rm
1556 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
1557 (memopv2i64 addr:$src2)))])>,
1562 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
1564 let isAsmParserOnly = 1 in {
1565 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr> {
1566 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
1567 !strconcat(OpcodeStr, "ps"), f256mem, [], [], 0>, VEX_4V;
1569 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
1570 !strconcat(OpcodeStr, "pd"), f256mem, [], [], 0>, OpSize, VEX_4V;
1574 // AVX 256-bit packed logical ops forms
1575 defm VAND : sse12_fp_packed_logical_y<0x54, "and">;
1576 defm VOR : sse12_fp_packed_logical_y<0x56, "or">;
1577 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor">;
1578 let isCommutable = 0 in
1579 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn">;
1581 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
1582 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
1583 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
1584 let isCommutable = 0 in
1585 defm ANDN : sse12_fp_packed_logical<0x55, "andn", undef /* dummy */, 1, [
1587 [(set VR128:$dst, (X86pandn VR128:$src1, VR128:$src2))],
1591 [(set VR128:$dst, (X86pandn VR128:$src1, (memopv2i64 addr:$src2)))],
1595 //===----------------------------------------------------------------------===//
1596 // SSE 1 & 2 - Arithmetic Instructions
1597 //===----------------------------------------------------------------------===//
1599 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
1602 /// In addition, we also have a special variant of the scalar form here to
1603 /// represent the associated intrinsic operation. This form is unlike the
1604 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
1605 /// and leaves the top elements unmodified (therefore these cannot be commuted).
1607 /// These three forms can each be reg+reg or reg+mem.
1610 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
1612 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
1614 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
1615 OpNode, FR32, f32mem, Is2Addr>, XS;
1616 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
1617 OpNode, FR64, f64mem, Is2Addr>, XD;
1620 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
1622 let mayLoad = 0 in {
1623 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
1624 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
1625 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
1626 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
1630 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
1632 let mayLoad = 0 in {
1633 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
1634 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
1635 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
1636 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
1640 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
1642 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1643 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
1644 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
1645 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
1648 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
1650 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1651 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
1652 SSEPackedSingle, Is2Addr>, TB;
1654 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
1655 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
1656 SSEPackedDouble, Is2Addr>, TB, OpSize;
1659 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
1660 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1661 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
1662 SSEPackedSingle, 0>, TB;
1664 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
1665 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
1666 SSEPackedDouble, 0>, TB, OpSize;
1669 // Binary Arithmetic instructions
1670 let isAsmParserOnly = 1 in {
1671 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
1672 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
1673 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
1674 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
1675 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
1676 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
1677 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
1678 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
1680 let isCommutable = 0 in {
1681 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
1682 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
1683 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
1684 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
1685 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
1686 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
1687 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
1688 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
1689 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
1690 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
1691 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
1692 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
1693 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
1694 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
1695 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
1696 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
1697 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
1698 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
1699 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
1700 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
1704 let Constraints = "$src1 = $dst" in {
1705 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
1706 basic_sse12_fp_binop_p<0x58, "add", fadd>,
1707 basic_sse12_fp_binop_s_int<0x58, "add">;
1708 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
1709 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
1710 basic_sse12_fp_binop_s_int<0x59, "mul">;
1712 let isCommutable = 0 in {
1713 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
1714 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
1715 basic_sse12_fp_binop_s_int<0x5C, "sub">;
1716 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
1717 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
1718 basic_sse12_fp_binop_s_int<0x5E, "div">;
1719 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
1720 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
1721 basic_sse12_fp_binop_s_int<0x5F, "max">,
1722 basic_sse12_fp_binop_p_int<0x5F, "max">;
1723 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
1724 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
1725 basic_sse12_fp_binop_s_int<0x5D, "min">,
1726 basic_sse12_fp_binop_p_int<0x5D, "min">;
1731 /// In addition, we also have a special variant of the scalar form here to
1732 /// represent the associated intrinsic operation. This form is unlike the
1733 /// plain scalar form, in that it takes an entire vector (instead of a
1734 /// scalar) and leaves the top elements undefined.
1736 /// And, we have a special variant form for a full-vector intrinsic form.
1738 /// sse1_fp_unop_s - SSE1 unops in scalar form.
1739 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
1740 SDNode OpNode, Intrinsic F32Int> {
1741 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
1742 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1743 [(set FR32:$dst, (OpNode FR32:$src))]>;
1744 // For scalar unary operations, fold a load into the operation
1745 // only in OptForSize mode. It eliminates an instruction, but it also
1746 // eliminates a whole-register clobber (the load), so it introduces a
1747 // partial register update condition.
1748 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
1749 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1750 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
1751 Requires<[HasSSE1, OptForSize]>;
1752 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1753 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1754 [(set VR128:$dst, (F32Int VR128:$src))]>;
1755 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1756 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
1757 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1760 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
1761 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1762 SDNode OpNode, Intrinsic F32Int> {
1763 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
1764 !strconcat(OpcodeStr,
1765 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1766 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2),
1767 !strconcat(OpcodeStr,
1768 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
1769 []>, XS, Requires<[HasAVX, OptForSize]>;
1770 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1771 !strconcat(OpcodeStr,
1772 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1773 [(set VR128:$dst, (F32Int VR128:$src))]>;
1774 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
1775 !strconcat(OpcodeStr,
1776 "ss\t{$src, $dst, $dst|$dst, $dst, $src}"),
1777 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
1780 /// sse1_fp_unop_p - SSE1 unops in packed form.
1781 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1782 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1783 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1784 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
1785 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1786 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1787 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
1790 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
1791 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1792 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1793 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1794 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
1795 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1796 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1797 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
1800 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
1801 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1802 Intrinsic V4F32Int> {
1803 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1804 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1805 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
1806 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1807 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1808 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
1811 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
1812 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1813 Intrinsic V4F32Int> {
1814 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1815 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1816 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
1817 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1818 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
1819 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
1822 /// sse2_fp_unop_s - SSE2 unops in scalar form.
1823 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
1824 SDNode OpNode, Intrinsic F64Int> {
1825 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
1826 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1827 [(set FR64:$dst, (OpNode FR64:$src))]>;
1828 // See the comments in sse1_fp_unop_s for why this is OptForSize.
1829 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
1830 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1831 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
1832 Requires<[HasSSE2, OptForSize]>;
1833 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1834 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1835 [(set VR128:$dst, (F64Int VR128:$src))]>;
1836 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1837 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
1838 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1841 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
1842 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr,
1843 SDNode OpNode, Intrinsic F64Int> {
1844 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
1845 !strconcat(OpcodeStr,
1846 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1847 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst),
1848 (ins FR64:$src1, f64mem:$src2),
1849 !strconcat(OpcodeStr,
1850 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
1851 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1852 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1853 [(set VR128:$dst, (F64Int VR128:$src))]>;
1854 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
1855 !strconcat(OpcodeStr, "sd\t{$src, $dst, $dst|$dst, $dst, $src}"),
1856 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
1859 /// sse2_fp_unop_p - SSE2 unops in vector forms.
1860 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
1862 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1863 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1864 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
1865 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1866 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1867 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
1870 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
1871 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
1872 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1873 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1874 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
1875 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1876 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1877 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
1880 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
1881 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
1882 Intrinsic V2F64Int> {
1883 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1884 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1885 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
1886 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1887 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1888 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
1891 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
1892 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
1893 Intrinsic V2F64Int> {
1894 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1895 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1896 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
1897 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1898 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
1899 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
1902 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
1904 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse_sqrt_ss>,
1905 sse2_fp_unop_s_avx<0x51, "vsqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1908 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
1909 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
1910 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1911 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
1912 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
1913 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
1914 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
1915 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
1918 // Reciprocal approximations. Note that these typically require refinement
1919 // in order to obtain suitable precision.
1920 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt", X86frsqrt,
1921 int_x86_sse_rsqrt_ss>, VEX_4V;
1922 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
1923 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
1924 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
1925 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
1927 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp", X86frcp, int_x86_sse_rcp_ss>,
1929 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
1930 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
1931 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
1932 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
1936 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
1937 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
1938 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
1939 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
1940 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
1941 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
1943 // Reciprocal approximations. Note that these typically require refinement
1944 // in order to obtain suitable precision.
1945 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
1946 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
1947 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
1948 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
1949 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
1950 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
1952 // There is no f64 version of the reciprocal approximation instructions.
1954 //===----------------------------------------------------------------------===//
1955 // SSE 1 & 2 - Non-temporal stores
1956 //===----------------------------------------------------------------------===//
1958 let isAsmParserOnly = 1 in {
1959 def VMOVNTPSmr_Int : VPSI<0x2B, MRMDestMem, (outs),
1960 (ins i128mem:$dst, VR128:$src),
1961 "movntps\t{$src, $dst|$dst, $src}",
1962 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>, VEX;
1963 def VMOVNTPDmr_Int : VPDI<0x2B, MRMDestMem, (outs),
1964 (ins i128mem:$dst, VR128:$src),
1965 "movntpd\t{$src, $dst|$dst, $src}",
1966 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>, VEX;
1968 let ExeDomain = SSEPackedInt in
1969 def VMOVNTDQmr_Int : VPDI<0xE7, MRMDestMem, (outs),
1970 (ins f128mem:$dst, VR128:$src),
1971 "movntdq\t{$src, $dst|$dst, $src}",
1972 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>, VEX;
1974 let AddedComplexity = 400 in { // Prefer non-temporal versions
1975 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
1976 (ins f128mem:$dst, VR128:$src),
1977 "movntps\t{$src, $dst|$dst, $src}",
1978 [(alignednontemporalstore (v4f32 VR128:$src),
1980 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
1981 (ins f128mem:$dst, VR128:$src),
1982 "movntpd\t{$src, $dst|$dst, $src}",
1983 [(alignednontemporalstore (v2f64 VR128:$src),
1985 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
1986 (ins f128mem:$dst, VR128:$src),
1987 "movntdq\t{$src, $dst|$dst, $src}",
1988 [(alignednontemporalstore (v2f64 VR128:$src),
1990 let ExeDomain = SSEPackedInt in
1991 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
1992 (ins f128mem:$dst, VR128:$src),
1993 "movntdq\t{$src, $dst|$dst, $src}",
1994 [(alignednontemporalstore (v4f32 VR128:$src),
1997 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
1998 (ins f256mem:$dst, VR256:$src),
1999 "movntps\t{$src, $dst|$dst, $src}",
2000 [(alignednontemporalstore (v8f32 VR256:$src),
2002 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2003 (ins f256mem:$dst, VR256:$src),
2004 "movntpd\t{$src, $dst|$dst, $src}",
2005 [(alignednontemporalstore (v4f64 VR256:$src),
2007 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2008 (ins f256mem:$dst, VR256:$src),
2009 "movntdq\t{$src, $dst|$dst, $src}",
2010 [(alignednontemporalstore (v4f64 VR256:$src),
2012 let ExeDomain = SSEPackedInt in
2013 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2014 (ins f256mem:$dst, VR256:$src),
2015 "movntdq\t{$src, $dst|$dst, $src}",
2016 [(alignednontemporalstore (v8f32 VR256:$src),
2021 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2022 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2023 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2024 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2025 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2026 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2028 def MOVNTPSmr_Int : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2029 "movntps\t{$src, $dst|$dst, $src}",
2030 [(int_x86_sse_movnt_ps addr:$dst, VR128:$src)]>;
2031 def MOVNTPDmr_Int : PDI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2032 "movntpd\t{$src, $dst|$dst, $src}",
2033 [(int_x86_sse2_movnt_pd addr:$dst, VR128:$src)]>;
2035 let ExeDomain = SSEPackedInt in
2036 def MOVNTDQmr_Int : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2037 "movntdq\t{$src, $dst|$dst, $src}",
2038 [(int_x86_sse2_movnt_dq addr:$dst, VR128:$src)]>;
2040 let AddedComplexity = 400 in { // Prefer non-temporal versions
2041 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2042 "movntps\t{$src, $dst|$dst, $src}",
2043 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2044 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2045 "movntpd\t{$src, $dst|$dst, $src}",
2046 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2048 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2049 "movntdq\t{$src, $dst|$dst, $src}",
2050 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2052 let ExeDomain = SSEPackedInt in
2053 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2054 "movntdq\t{$src, $dst|$dst, $src}",
2055 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2057 // There is no AVX form for instructions below this point
2058 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2059 "movnti\t{$src, $dst|$dst, $src}",
2060 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2061 TB, Requires<[HasSSE2]>;
2063 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2064 "movnti\t{$src, $dst|$dst, $src}",
2065 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2066 TB, Requires<[HasSSE2]>;
2069 def MOVNTImr_Int : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2070 "movnti\t{$src, $dst|$dst, $src}",
2071 [(int_x86_sse2_movnt_i addr:$dst, GR32:$src)]>,
2072 TB, Requires<[HasSSE2]>;
2074 //===----------------------------------------------------------------------===//
2075 // SSE 1 & 2 - Misc Instructions (No AVX form)
2076 //===----------------------------------------------------------------------===//
2078 // Prefetch intrinsic.
2079 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2080 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
2081 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2082 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
2083 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2084 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
2085 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2086 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
2088 // Load, store, and memory fence
2089 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2090 TB, Requires<[HasSSE1]>;
2091 def : Pat<(X86SFence), (SFENCE)>;
2093 // Alias instructions that map zero vector to pxor / xorp* for sse.
2094 // We set canFoldAsLoad because this can be converted to a constant-pool
2095 // load of an all-zeros value if folding it would be beneficial.
2096 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2097 // JIT implementation, it does not expand the instructions below like
2098 // X86MCInstLower does.
2099 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2100 isCodeGenOnly = 1 in {
2101 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2102 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
2103 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2104 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
2105 let ExeDomain = SSEPackedInt in
2106 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2107 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2110 // The same as done above but for AVX. The 128-bit versions are the
2111 // same, but re-encoded. The 256-bit does not support PI version.
2112 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
2113 // JIT implementatioan, it does not expand the instructions below like
2114 // X86MCInstLower does.
2115 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
2116 isCodeGenOnly = 1, Predicates = [HasAVX] in {
2117 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2118 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
2119 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
2120 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
2121 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2122 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
2123 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
2124 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
2125 let ExeDomain = SSEPackedInt in
2126 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
2127 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
2130 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
2131 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
2132 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
2134 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
2135 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
2137 //===----------------------------------------------------------------------===//
2138 // SSE 1 & 2 - Load/Store XCSR register
2139 //===----------------------------------------------------------------------===//
2141 let isAsmParserOnly = 1 in {
2142 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2143 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2144 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2145 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2148 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2149 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2150 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2151 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2153 //===---------------------------------------------------------------------===//
2154 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2155 //===---------------------------------------------------------------------===//
2157 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2159 let isAsmParserOnly = 1 in {
2160 let neverHasSideEffects = 1 in {
2161 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2162 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2163 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2164 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2166 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2167 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2168 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2169 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2171 let canFoldAsLoad = 1, mayLoad = 1 in {
2172 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2173 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2174 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2175 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2176 let Predicates = [HasAVX] in {
2177 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2178 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2179 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2180 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2184 let mayStore = 1 in {
2185 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2186 (ins i128mem:$dst, VR128:$src),
2187 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2188 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2189 (ins i256mem:$dst, VR256:$src),
2190 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2191 let Predicates = [HasAVX] in {
2192 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2193 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2194 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2195 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2200 let neverHasSideEffects = 1 in
2201 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2202 "movdqa\t{$src, $dst|$dst, $src}", []>;
2204 let canFoldAsLoad = 1, mayLoad = 1 in {
2205 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2206 "movdqa\t{$src, $dst|$dst, $src}",
2207 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2208 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2209 "movdqu\t{$src, $dst|$dst, $src}",
2210 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2211 XS, Requires<[HasSSE2]>;
2214 let mayStore = 1 in {
2215 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2216 "movdqa\t{$src, $dst|$dst, $src}",
2217 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2218 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2219 "movdqu\t{$src, $dst|$dst, $src}",
2220 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2221 XS, Requires<[HasSSE2]>;
2224 // Intrinsic forms of MOVDQU load and store
2225 let isAsmParserOnly = 1 in {
2226 let canFoldAsLoad = 1 in
2227 def VMOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2228 "vmovdqu\t{$src, $dst|$dst, $src}",
2229 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2230 XS, VEX, Requires<[HasAVX]>;
2231 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2232 "vmovdqu\t{$src, $dst|$dst, $src}",
2233 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2234 XS, VEX, Requires<[HasAVX]>;
2237 let canFoldAsLoad = 1 in
2238 def MOVDQUrm_Int : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2239 "movdqu\t{$src, $dst|$dst, $src}",
2240 [(set VR128:$dst, (int_x86_sse2_loadu_dq addr:$src))]>,
2241 XS, Requires<[HasSSE2]>;
2242 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2243 "movdqu\t{$src, $dst|$dst, $src}",
2244 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2245 XS, Requires<[HasSSE2]>;
2247 } // ExeDomain = SSEPackedInt
2249 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2250 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2251 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2253 //===---------------------------------------------------------------------===//
2254 // SSE2 - Packed Integer Arithmetic Instructions
2255 //===---------------------------------------------------------------------===//
2257 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2259 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2260 bit IsCommutable = 0, bit Is2Addr = 1> {
2261 let isCommutable = IsCommutable in
2262 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2263 (ins VR128:$src1, VR128:$src2),
2265 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2266 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2267 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2268 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2269 (ins VR128:$src1, i128mem:$src2),
2271 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2272 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2273 [(set VR128:$dst, (IntId VR128:$src1,
2274 (bitconvert (memopv2i64 addr:$src2))))]>;
2277 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2278 string OpcodeStr, Intrinsic IntId,
2279 Intrinsic IntId2, bit Is2Addr = 1> {
2280 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2281 (ins VR128:$src1, VR128:$src2),
2283 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2284 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2285 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2286 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2287 (ins VR128:$src1, i128mem:$src2),
2289 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2290 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2291 [(set VR128:$dst, (IntId VR128:$src1,
2292 (bitconvert (memopv2i64 addr:$src2))))]>;
2293 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2294 (ins VR128:$src1, i32i8imm:$src2),
2296 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2297 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2298 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2301 /// PDI_binop_rm - Simple SSE2 binary operator.
2302 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2303 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2304 let isCommutable = IsCommutable in
2305 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2306 (ins VR128:$src1, VR128:$src2),
2308 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2309 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2310 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2311 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2312 (ins VR128:$src1, i128mem:$src2),
2314 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2316 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2317 (bitconvert (memopv2i64 addr:$src2)))))]>;
2320 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2322 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2323 /// to collapse (bitconvert VT to VT) into its operand.
2325 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2326 bit IsCommutable = 0, bit Is2Addr = 1> {
2327 let isCommutable = IsCommutable in
2328 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2329 (ins VR128:$src1, VR128:$src2),
2331 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2332 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2333 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2334 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2335 (ins VR128:$src1, i128mem:$src2),
2337 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2339 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2342 } // ExeDomain = SSEPackedInt
2344 // 128-bit Integer Arithmetic
2346 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2347 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2348 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2349 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2350 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2351 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2352 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2353 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2354 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2355 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2358 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2360 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2362 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2364 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2366 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2368 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2370 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2372 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2374 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2376 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2378 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2380 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2382 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2384 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2386 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2388 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2390 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2392 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2394 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2398 let Constraints = "$src1 = $dst" in {
2399 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2400 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2401 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2402 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2403 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2404 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2405 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2406 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2407 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2410 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2411 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2412 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2413 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2414 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2415 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2416 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2417 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2418 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2419 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2420 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2421 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2422 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2423 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2424 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2425 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2426 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2427 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2428 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2430 } // Constraints = "$src1 = $dst"
2432 //===---------------------------------------------------------------------===//
2433 // SSE2 - Packed Integer Logical Instructions
2434 //===---------------------------------------------------------------------===//
2436 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2437 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2438 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2440 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2441 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2443 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2444 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2447 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2448 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2450 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2451 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2453 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2454 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2457 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2458 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2460 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2461 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2464 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2465 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2466 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2468 let ExeDomain = SSEPackedInt in {
2469 let neverHasSideEffects = 1 in {
2470 // 128-bit logical shifts.
2471 def VPSLLDQri : PDIi8<0x73, MRM7r,
2472 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2473 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2475 def VPSRLDQri : PDIi8<0x73, MRM3r,
2476 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2477 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2479 // PSRADQri doesn't exist in SSE[1-3].
2481 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2482 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2483 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2484 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2485 VR128:$src2)))]>, VEX_4V;
2487 def VPANDNrm : PDI<0xDF, MRMSrcMem,
2488 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2489 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2490 [(set VR128:$dst, (v2i64 (and (vnot VR128:$src1),
2491 (memopv2i64 addr:$src2))))]>,
2496 let Constraints = "$src1 = $dst" in {
2497 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
2498 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
2499 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
2500 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
2501 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
2502 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
2504 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
2505 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
2506 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
2507 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
2508 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
2509 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
2511 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
2512 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
2513 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
2514 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
2516 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
2517 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
2518 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
2520 let ExeDomain = SSEPackedInt in {
2521 let neverHasSideEffects = 1 in {
2522 // 128-bit logical shifts.
2523 def PSLLDQri : PDIi8<0x73, MRM7r,
2524 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2525 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
2526 def PSRLDQri : PDIi8<0x73, MRM3r,
2527 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2528 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
2529 // PSRADQri doesn't exist in SSE[1-3].
2531 def PANDNrr : PDI<0xDF, MRMSrcReg,
2532 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2533 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2535 def PANDNrm : PDI<0xDF, MRMSrcMem,
2536 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2537 "pandn\t{$src2, $dst|$dst, $src2}", []>;
2539 } // Constraints = "$src1 = $dst"
2541 let Predicates = [HasAVX] in {
2542 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2543 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2544 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2545 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2546 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2547 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
2548 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2549 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
2550 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2551 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2553 // Shift up / down and insert zero's.
2554 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2555 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2556 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2557 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2560 let Predicates = [HasSSE2] in {
2561 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
2562 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2563 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
2564 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2565 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
2566 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
2567 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
2568 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
2569 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
2570 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
2572 // Shift up / down and insert zero's.
2573 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
2574 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2575 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
2576 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
2579 //===---------------------------------------------------------------------===//
2580 // SSE2 - Packed Integer Comparison Instructions
2581 //===---------------------------------------------------------------------===//
2583 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2584 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
2586 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
2588 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
2590 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
2592 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
2594 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
2598 let Constraints = "$src1 = $dst" in {
2599 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
2600 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
2601 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
2602 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
2603 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
2604 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
2605 } // Constraints = "$src1 = $dst"
2607 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
2608 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
2609 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
2610 (PCMPEQBrm VR128:$src1, addr:$src2)>;
2611 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
2612 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
2613 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
2614 (PCMPEQWrm VR128:$src1, addr:$src2)>;
2615 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
2616 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
2617 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
2618 (PCMPEQDrm VR128:$src1, addr:$src2)>;
2620 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
2621 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
2622 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
2623 (PCMPGTBrm VR128:$src1, addr:$src2)>;
2624 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
2625 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
2626 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
2627 (PCMPGTWrm VR128:$src1, addr:$src2)>;
2628 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
2629 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
2630 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
2631 (PCMPGTDrm VR128:$src1, addr:$src2)>;
2633 //===---------------------------------------------------------------------===//
2634 // SSE2 - Packed Integer Pack Instructions
2635 //===---------------------------------------------------------------------===//
2637 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2638 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
2640 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
2642 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
2646 let Constraints = "$src1 = $dst" in {
2647 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
2648 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
2649 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
2650 } // Constraints = "$src1 = $dst"
2652 //===---------------------------------------------------------------------===//
2653 // SSE2 - Packed Integer Shuffle Instructions
2654 //===---------------------------------------------------------------------===//
2656 let ExeDomain = SSEPackedInt in {
2657 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
2659 def ri : Ii8<0x70, MRMSrcReg,
2660 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
2661 !strconcat(OpcodeStr,
2662 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2663 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
2665 def mi : Ii8<0x70, MRMSrcMem,
2666 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
2667 !strconcat(OpcodeStr,
2668 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
2669 [(set VR128:$dst, (vt (pshuf_frag:$src2
2670 (bc_frag (memopv2i64 addr:$src1)),
2673 } // ExeDomain = SSEPackedInt
2675 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2676 let AddedComplexity = 5 in
2677 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, OpSize,
2680 // SSE2 with ImmT == Imm8 and XS prefix.
2681 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
2684 // SSE2 with ImmT == Imm8 and XD prefix.
2685 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
2689 let Predicates = [HasSSE2] in {
2690 let AddedComplexity = 5 in
2691 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
2693 // SSE2 with ImmT == Imm8 and XS prefix.
2694 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
2696 // SSE2 with ImmT == Imm8 and XD prefix.
2697 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
2700 //===---------------------------------------------------------------------===//
2701 // SSE2 - Packed Integer Unpack Instructions
2702 //===---------------------------------------------------------------------===//
2704 let ExeDomain = SSEPackedInt in {
2705 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
2706 PatFrag unp_frag, PatFrag bc_frag, bit Is2Addr = 1> {
2707 def rr : PDI<opc, MRMSrcReg,
2708 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2710 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2711 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2712 [(set VR128:$dst, (vt (unp_frag VR128:$src1, VR128:$src2)))]>;
2713 def rm : PDI<opc, MRMSrcMem,
2714 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2716 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
2717 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2718 [(set VR128:$dst, (unp_frag VR128:$src1,
2719 (bc_frag (memopv2i64
2723 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2724 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, unpckl, bc_v16i8,
2726 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, unpckl, bc_v8i16,
2728 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, unpckl, bc_v4i32,
2731 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2732 /// knew to collapse (bitconvert VT to VT) into its operand.
2733 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2734 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2735 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2737 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>, VEX_4V;
2738 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2739 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2740 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2742 (v2i64 (unpckl VR128:$src1,
2743 (memopv2i64 addr:$src2))))]>, VEX_4V;
2745 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, unpckh, bc_v16i8,
2747 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, unpckh, bc_v8i16,
2749 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, unpckh, bc_v4i32,
2752 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2753 /// knew to collapse (bitconvert VT to VT) into its operand.
2754 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2755 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2756 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2758 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>, VEX_4V;
2759 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2760 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2761 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2763 (v2i64 (unpckh VR128:$src1,
2764 (memopv2i64 addr:$src2))))]>, VEX_4V;
2767 let Constraints = "$src1 = $dst" in {
2768 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, unpckl, bc_v16i8>;
2769 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, unpckl, bc_v8i16>;
2770 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, unpckl, bc_v4i32>;
2772 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2773 /// knew to collapse (bitconvert VT to VT) into its operand.
2774 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
2775 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2776 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2778 (v2i64 (unpckl VR128:$src1, VR128:$src2)))]>;
2779 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
2780 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2781 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
2783 (v2i64 (unpckl VR128:$src1,
2784 (memopv2i64 addr:$src2))))]>;
2786 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, unpckh, bc_v16i8>;
2787 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, unpckh, bc_v8i16>;
2788 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, unpckh, bc_v4i32>;
2790 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
2791 /// knew to collapse (bitconvert VT to VT) into its operand.
2792 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
2793 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2794 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2796 (v2i64 (unpckh VR128:$src1, VR128:$src2)))]>;
2797 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
2798 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
2799 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
2801 (v2i64 (unpckh VR128:$src1,
2802 (memopv2i64 addr:$src2))))]>;
2805 } // ExeDomain = SSEPackedInt
2807 //===---------------------------------------------------------------------===//
2808 // SSE2 - Packed Integer Extract and Insert
2809 //===---------------------------------------------------------------------===//
2811 let ExeDomain = SSEPackedInt in {
2812 multiclass sse2_pinsrw<bit Is2Addr = 1> {
2813 def rri : Ii8<0xC4, MRMSrcReg,
2814 (outs VR128:$dst), (ins VR128:$src1,
2815 GR32:$src2, i32i8imm:$src3),
2817 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2818 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
2820 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
2821 def rmi : Ii8<0xC4, MRMSrcMem,
2822 (outs VR128:$dst), (ins VR128:$src1,
2823 i16mem:$src2, i32i8imm:$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, (extloadi16 addr:$src2),
2833 let isAsmParserOnly = 1, Predicates = [HasAVX] in
2834 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
2835 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2836 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2837 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2838 imm:$src2))]>, OpSize, VEX;
2839 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
2840 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
2841 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2842 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
2846 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
2847 defm VPINSRW : sse2_pinsrw<0>, OpSize, VEX_4V;
2848 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
2849 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
2850 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2851 []>, OpSize, VEX_4V;
2854 let Constraints = "$src1 = $dst" in
2855 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
2857 } // ExeDomain = SSEPackedInt
2859 //===---------------------------------------------------------------------===//
2860 // SSE2 - Packed Mask Creation
2861 //===---------------------------------------------------------------------===//
2863 let ExeDomain = SSEPackedInt in {
2865 let isAsmParserOnly = 1 in {
2866 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2867 "pmovmskb\t{$src, $dst|$dst, $src}",
2868 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
2869 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2870 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
2872 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
2873 "pmovmskb\t{$src, $dst|$dst, $src}",
2874 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
2876 } // ExeDomain = SSEPackedInt
2878 //===---------------------------------------------------------------------===//
2879 // SSE2 - Conditional Store
2880 //===---------------------------------------------------------------------===//
2882 let ExeDomain = SSEPackedInt in {
2884 let isAsmParserOnly = 1 in {
2886 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
2887 (ins VR128:$src, VR128:$mask),
2888 "maskmovdqu\t{$mask, $src|$src, $mask}",
2889 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
2891 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
2892 (ins VR128:$src, VR128:$mask),
2893 "maskmovdqu\t{$mask, $src|$src, $mask}",
2894 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
2898 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2899 "maskmovdqu\t{$mask, $src|$src, $mask}",
2900 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
2902 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
2903 "maskmovdqu\t{$mask, $src|$src, $mask}",
2904 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
2906 } // ExeDomain = SSEPackedInt
2908 //===---------------------------------------------------------------------===//
2909 // SSE2 - Move Doubleword
2910 //===---------------------------------------------------------------------===//
2912 // Move Int Doubleword to Packed Double Int
2913 let isAsmParserOnly = 1 in {
2914 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2915 "movd\t{$src, $dst|$dst, $src}",
2917 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
2918 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2919 "movd\t{$src, $dst|$dst, $src}",
2921 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
2924 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
2925 "movd\t{$src, $dst|$dst, $src}",
2927 (v4i32 (scalar_to_vector GR32:$src)))]>;
2928 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
2929 "movd\t{$src, $dst|$dst, $src}",
2931 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
2932 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
2933 "mov{d|q}\t{$src, $dst|$dst, $src}",
2935 (v2i64 (scalar_to_vector GR64:$src)))]>;
2936 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
2937 "mov{d|q}\t{$src, $dst|$dst, $src}",
2938 [(set FR64:$dst, (bitconvert GR64:$src))]>;
2941 // Move Int Doubleword to Single Scalar
2942 let isAsmParserOnly = 1 in {
2943 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2944 "movd\t{$src, $dst|$dst, $src}",
2945 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
2947 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2948 "movd\t{$src, $dst|$dst, $src}",
2949 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
2952 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
2953 "movd\t{$src, $dst|$dst, $src}",
2954 [(set FR32:$dst, (bitconvert GR32:$src))]>;
2956 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
2957 "movd\t{$src, $dst|$dst, $src}",
2958 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
2960 // Move Packed Doubleword Int to Packed Double Int
2961 let isAsmParserOnly = 1 in {
2962 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2963 "movd\t{$src, $dst|$dst, $src}",
2964 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2966 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
2967 (ins i32mem:$dst, VR128:$src),
2968 "movd\t{$src, $dst|$dst, $src}",
2969 [(store (i32 (vector_extract (v4i32 VR128:$src),
2970 (iPTR 0))), addr:$dst)]>, VEX;
2972 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
2973 "movd\t{$src, $dst|$dst, $src}",
2974 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
2976 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
2977 "movd\t{$src, $dst|$dst, $src}",
2978 [(store (i32 (vector_extract (v4i32 VR128:$src),
2979 (iPTR 0))), addr:$dst)]>;
2981 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
2982 "mov{d|q}\t{$src, $dst|$dst, $src}",
2983 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
2985 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
2986 "movq\t{$src, $dst|$dst, $src}",
2987 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
2989 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
2990 "mov{d|q}\t{$src, $dst|$dst, $src}",
2991 [(set GR64:$dst, (bitconvert FR64:$src))]>;
2992 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
2993 "movq\t{$src, $dst|$dst, $src}",
2994 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
2996 // Move Scalar Single to Double Int
2997 let isAsmParserOnly = 1 in {
2998 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
2999 "movd\t{$src, $dst|$dst, $src}",
3000 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3001 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3002 "movd\t{$src, $dst|$dst, $src}",
3003 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3005 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3006 "movd\t{$src, $dst|$dst, $src}",
3007 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3008 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3009 "movd\t{$src, $dst|$dst, $src}",
3010 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3012 // movd / movq to XMM register zero-extends
3013 let AddedComplexity = 15, isAsmParserOnly = 1 in {
3014 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3015 "movd\t{$src, $dst|$dst, $src}",
3016 [(set VR128:$dst, (v4i32 (X86vzmovl
3017 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3019 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3020 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3021 [(set VR128:$dst, (v2i64 (X86vzmovl
3022 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3025 let AddedComplexity = 15 in {
3026 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3027 "movd\t{$src, $dst|$dst, $src}",
3028 [(set VR128:$dst, (v4i32 (X86vzmovl
3029 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3030 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3031 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3032 [(set VR128:$dst, (v2i64 (X86vzmovl
3033 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3036 let AddedComplexity = 20 in {
3037 let isAsmParserOnly = 1 in
3038 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3039 "movd\t{$src, $dst|$dst, $src}",
3041 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3042 (loadi32 addr:$src))))))]>,
3044 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3045 "movd\t{$src, $dst|$dst, $src}",
3047 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3048 (loadi32 addr:$src))))))]>;
3050 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3051 (MOVZDI2PDIrm addr:$src)>;
3052 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3053 (MOVZDI2PDIrm addr:$src)>;
3054 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3055 (MOVZDI2PDIrm addr:$src)>;
3058 //===---------------------------------------------------------------------===//
3059 // SSE2 - Move Quadword
3060 //===---------------------------------------------------------------------===//
3062 // Move Quadword Int to Packed Quadword Int
3063 let isAsmParserOnly = 1 in
3064 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3065 "vmovq\t{$src, $dst|$dst, $src}",
3067 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3068 VEX, Requires<[HasAVX]>;
3069 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3070 "movq\t{$src, $dst|$dst, $src}",
3072 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3073 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3075 // Move Packed Quadword Int to Quadword Int
3076 let isAsmParserOnly = 1 in
3077 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3078 "movq\t{$src, $dst|$dst, $src}",
3079 [(store (i64 (vector_extract (v2i64 VR128:$src),
3080 (iPTR 0))), addr:$dst)]>, VEX;
3081 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3082 "movq\t{$src, $dst|$dst, $src}",
3083 [(store (i64 (vector_extract (v2i64 VR128:$src),
3084 (iPTR 0))), addr:$dst)]>;
3086 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3087 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3089 // Store / copy lower 64-bits of a XMM register.
3090 let isAsmParserOnly = 1 in
3091 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3092 "movq\t{$src, $dst|$dst, $src}",
3093 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3094 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3095 "movq\t{$src, $dst|$dst, $src}",
3096 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3098 let AddedComplexity = 20, isAsmParserOnly = 1 in
3099 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3100 "vmovq\t{$src, $dst|$dst, $src}",
3102 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3103 (loadi64 addr:$src))))))]>,
3104 XS, VEX, Requires<[HasAVX]>;
3106 let AddedComplexity = 20 in {
3107 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3108 "movq\t{$src, $dst|$dst, $src}",
3110 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3111 (loadi64 addr:$src))))))]>,
3112 XS, Requires<[HasSSE2]>;
3114 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3115 (MOVZQI2PQIrm addr:$src)>;
3116 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3117 (MOVZQI2PQIrm addr:$src)>;
3118 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3121 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3122 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3123 let isAsmParserOnly = 1, AddedComplexity = 15 in
3124 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3125 "vmovq\t{$src, $dst|$dst, $src}",
3126 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3127 XS, VEX, Requires<[HasAVX]>;
3128 let AddedComplexity = 15 in
3129 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3130 "movq\t{$src, $dst|$dst, $src}",
3131 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3132 XS, Requires<[HasSSE2]>;
3134 let AddedComplexity = 20, isAsmParserOnly = 1 in
3135 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3136 "vmovq\t{$src, $dst|$dst, $src}",
3137 [(set VR128:$dst, (v2i64 (X86vzmovl
3138 (loadv2i64 addr:$src))))]>,
3139 XS, VEX, Requires<[HasAVX]>;
3140 let AddedComplexity = 20 in {
3141 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3142 "movq\t{$src, $dst|$dst, $src}",
3143 [(set VR128:$dst, (v2i64 (X86vzmovl
3144 (loadv2i64 addr:$src))))]>,
3145 XS, Requires<[HasSSE2]>;
3147 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3148 (MOVZPQILo2PQIrm addr:$src)>;
3151 // Instructions to match in the assembler
3152 let isAsmParserOnly = 1 in {
3153 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3154 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3155 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3156 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3157 // Recognize "movd" with GR64 destination, but encode as a "movq"
3158 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3159 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3162 // Instructions for the disassembler
3163 // xr = XMM register
3166 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3167 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3168 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3169 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3170 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3172 //===---------------------------------------------------------------------===//
3173 // SSE2 - Misc Instructions
3174 //===---------------------------------------------------------------------===//
3177 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3178 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3179 TB, Requires<[HasSSE2]>;
3181 // Load, store, and memory fence
3182 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3183 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3184 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3185 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3186 def : Pat<(X86LFence), (LFENCE)>;
3187 def : Pat<(X86MFence), (MFENCE)>;
3190 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3191 // was introduced with SSE2, it's backward compatible.
3192 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3194 // Alias instructions that map zero vector to pxor / xorp* for sse.
3195 // We set canFoldAsLoad because this can be converted to a constant-pool
3196 // load of an all-ones value if folding it would be beneficial.
3197 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3198 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3199 // FIXME: Change encoding to pseudo.
3200 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3201 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3203 //===---------------------------------------------------------------------===//
3204 // SSE3 - Conversion Instructions
3205 //===---------------------------------------------------------------------===//
3207 // Convert Packed Double FP to Packed DW Integers
3208 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3209 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3210 // register, but the same isn't true when using memory operands instead.
3211 // Provide other assembly rr and rm forms to address this explicitly.
3212 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3213 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3214 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3215 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3218 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3219 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3220 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3221 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3224 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3225 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3226 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3227 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3230 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3231 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3232 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3233 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3235 // Convert Packed DW Integers to Packed Double FP
3236 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3237 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3238 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3239 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3240 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3241 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3242 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3243 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3244 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3247 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3248 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3249 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3250 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3252 // AVX 256-bit register conversion intrinsics
3253 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3254 (VCVTDQ2PDYrr VR128:$src)>;
3255 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3256 (VCVTDQ2PDYrm addr:$src)>;
3258 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3259 (VCVTPD2DQYrr VR256:$src)>;
3260 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3261 (VCVTPD2DQYrm addr:$src)>;
3263 //===---------------------------------------------------------------------===//
3264 // SSE3 - Move Instructions
3265 //===---------------------------------------------------------------------===//
3267 // Replicate Single FP
3268 multiclass sse3_replicate_sfp<bits<8> op, PatFrag rep_frag, string OpcodeStr> {
3269 def rr : S3SI<op, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3270 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3271 [(set VR128:$dst, (v4f32 (rep_frag
3272 VR128:$src, (undef))))]>;
3273 def rm : S3SI<op, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3274 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3275 [(set VR128:$dst, (rep_frag
3276 (memopv4f32 addr:$src), (undef)))]>;
3279 multiclass sse3_replicate_sfp_y<bits<8> op, PatFrag rep_frag,
3281 def rr : S3SI<op, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3282 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3283 def rm : S3SI<op, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3284 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"), []>;
3287 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3288 // FIXME: Merge above classes when we have patterns for the ymm version
3289 defm VMOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "vmovshdup">, VEX;
3290 defm VMOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "vmovsldup">, VEX;
3291 defm VMOVSHDUPY : sse3_replicate_sfp_y<0x16, movshdup, "vmovshdup">, VEX;
3292 defm VMOVSLDUPY : sse3_replicate_sfp_y<0x12, movsldup, "vmovsldup">, VEX;
3294 defm MOVSHDUP : sse3_replicate_sfp<0x16, movshdup, "movshdup">;
3295 defm MOVSLDUP : sse3_replicate_sfp<0x12, movsldup, "movsldup">;
3297 // Replicate Double FP
3298 multiclass sse3_replicate_dfp<string OpcodeStr> {
3299 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3300 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3301 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3302 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3303 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3305 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3309 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3310 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3311 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3313 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3314 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3318 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3319 // FIXME: Merge above classes when we have patterns for the ymm version
3320 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3321 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3323 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3325 // Move Unaligned Integer
3326 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3327 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3328 "vlddqu\t{$src, $dst|$dst, $src}",
3329 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
3330 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3331 "vlddqu\t{$src, $dst|$dst, $src}",
3332 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
3334 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3335 "lddqu\t{$src, $dst|$dst, $src}",
3336 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
3338 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3340 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3342 // Several Move patterns
3343 let AddedComplexity = 5 in {
3344 def : Pat<(movddup (memopv2f64 addr:$src), (undef)),
3345 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3346 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
3347 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3348 def : Pat<(movddup (memopv2i64 addr:$src), (undef)),
3349 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3350 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
3351 (MOVDDUPrm addr:$src)>, Requires<[HasSSE3]>;
3354 // vector_shuffle v1, <undef> <1, 1, 3, 3>
3355 let AddedComplexity = 15 in
3356 def : Pat<(v4i32 (movshdup VR128:$src, (undef))),
3357 (MOVSHDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3358 let AddedComplexity = 20 in
3359 def : Pat<(v4i32 (movshdup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3360 (MOVSHDUPrm addr:$src)>, Requires<[HasSSE3]>;
3362 // vector_shuffle v1, <undef> <0, 0, 2, 2>
3363 let AddedComplexity = 15 in
3364 def : Pat<(v4i32 (movsldup VR128:$src, (undef))),
3365 (MOVSLDUPrr VR128:$src)>, Requires<[HasSSE3]>;
3366 let AddedComplexity = 20 in
3367 def : Pat<(v4i32 (movsldup (bc_v4i32 (memopv2i64 addr:$src)), (undef))),
3368 (MOVSLDUPrm addr:$src)>, Requires<[HasSSE3]>;
3370 //===---------------------------------------------------------------------===//
3371 // SSE3 - Arithmetic
3372 //===---------------------------------------------------------------------===//
3374 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
3375 X86MemOperand x86memop, bit Is2Addr = 1> {
3376 def rr : I<0xD0, MRMSrcReg,
3377 (outs RC:$dst), (ins RC:$src1, RC:$src2),
3379 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3380 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3381 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
3382 def rm : I<0xD0, MRMSrcMem,
3383 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3385 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3386 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3387 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
3390 let isAsmParserOnly = 1, Predicates = [HasAVX],
3391 ExeDomain = SSEPackedDouble in {
3392 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
3393 f128mem, 0>, XD, VEX_4V;
3394 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
3395 f128mem, 0>, OpSize, VEX_4V;
3396 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
3397 f256mem, 0>, XD, VEX_4V;
3398 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
3399 f256mem, 0>, OpSize, VEX_4V;
3401 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
3402 ExeDomain = SSEPackedDouble in {
3403 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
3405 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
3406 f128mem>, TB, OpSize;
3409 //===---------------------------------------------------------------------===//
3410 // SSE3 Instructions
3411 //===---------------------------------------------------------------------===//
3414 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3415 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3416 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3418 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3419 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3420 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3422 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3424 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3425 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3426 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3428 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
3429 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
3430 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
3432 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3433 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3434 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
3436 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
3438 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3439 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3440 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
3443 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3444 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
3445 int_x86_sse3_hadd_ps, 0>, VEX_4V;
3446 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
3447 int_x86_sse3_hadd_pd, 0>, VEX_4V;
3448 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
3449 int_x86_sse3_hsub_ps, 0>, VEX_4V;
3450 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
3451 int_x86_sse3_hsub_pd, 0>, VEX_4V;
3452 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
3453 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
3454 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
3455 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
3456 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
3457 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
3458 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
3459 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
3462 let Constraints = "$src1 = $dst" in {
3463 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
3464 int_x86_sse3_hadd_ps>;
3465 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
3466 int_x86_sse3_hadd_pd>;
3467 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
3468 int_x86_sse3_hsub_ps>;
3469 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
3470 int_x86_sse3_hsub_pd>;
3473 //===---------------------------------------------------------------------===//
3474 // SSSE3 - Packed Absolute Instructions
3475 //===---------------------------------------------------------------------===//
3478 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
3479 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
3480 PatFrag mem_frag128, Intrinsic IntId128> {
3481 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3483 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3484 [(set VR128:$dst, (IntId128 VR128:$src))]>,
3487 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3489 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3492 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
3495 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3496 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
3497 int_x86_ssse3_pabs_b_128>, VEX;
3498 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
3499 int_x86_ssse3_pabs_w_128>, VEX;
3500 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
3501 int_x86_ssse3_pabs_d_128>, VEX;
3504 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
3505 int_x86_ssse3_pabs_b_128>;
3506 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
3507 int_x86_ssse3_pabs_w_128>;
3508 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
3509 int_x86_ssse3_pabs_d_128>;
3511 //===---------------------------------------------------------------------===//
3512 // SSSE3 - Packed Binary Operator Instructions
3513 //===---------------------------------------------------------------------===//
3515 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
3516 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
3517 PatFrag mem_frag128, Intrinsic IntId128,
3519 let isCommutable = 1 in
3520 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
3521 (ins VR128:$src1, VR128:$src2),
3523 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3524 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3525 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
3527 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
3528 (ins VR128:$src1, i128mem:$src2),
3530 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3531 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3533 (IntId128 VR128:$src1,
3534 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
3537 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3538 let isCommutable = 0 in {
3539 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
3540 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
3541 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
3542 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
3543 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
3544 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
3545 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
3546 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
3547 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
3548 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
3549 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
3550 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
3551 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
3552 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
3553 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
3554 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
3555 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
3556 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
3557 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
3558 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
3559 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
3560 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
3562 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
3563 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
3566 // None of these have i8 immediate fields.
3567 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
3568 let isCommutable = 0 in {
3569 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
3570 int_x86_ssse3_phadd_w_128>;
3571 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
3572 int_x86_ssse3_phadd_d_128>;
3573 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
3574 int_x86_ssse3_phadd_sw_128>;
3575 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
3576 int_x86_ssse3_phsub_w_128>;
3577 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
3578 int_x86_ssse3_phsub_d_128>;
3579 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
3580 int_x86_ssse3_phsub_sw_128>;
3581 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
3582 int_x86_ssse3_pmadd_ub_sw_128>;
3583 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
3584 int_x86_ssse3_pshuf_b_128>;
3585 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
3586 int_x86_ssse3_psign_b_128>;
3587 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
3588 int_x86_ssse3_psign_w_128>;
3589 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
3590 int_x86_ssse3_psign_d_128>;
3592 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
3593 int_x86_ssse3_pmul_hr_sw_128>;
3596 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
3597 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
3598 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
3599 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
3601 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
3602 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3603 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
3604 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3605 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
3606 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
3608 //===---------------------------------------------------------------------===//
3609 // SSSE3 - Packed Align Instruction Patterns
3610 //===---------------------------------------------------------------------===//
3612 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
3613 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
3614 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
3616 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3618 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3620 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
3621 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
3623 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
3625 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
3629 let isAsmParserOnly = 1, Predicates = [HasAVX] in
3630 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
3631 let Constraints = "$src1 = $dst" in
3632 defm PALIGN : ssse3_palign<"palignr">;
3634 let AddedComplexity = 5 in {
3635 def : Pat<(v4i32 (palign:$src3 VR128:$src1, VR128:$src2)),
3636 (PALIGNR128rr VR128:$src2, VR128:$src1,
3637 (SHUFFLE_get_palign_imm VR128:$src3))>,
3638 Requires<[HasSSSE3]>;
3639 def : Pat<(v4f32 (palign:$src3 VR128:$src1, VR128:$src2)),
3640 (PALIGNR128rr VR128:$src2, VR128:$src1,
3641 (SHUFFLE_get_palign_imm VR128:$src3))>,
3642 Requires<[HasSSSE3]>;
3643 def : Pat<(v8i16 (palign:$src3 VR128:$src1, VR128:$src2)),
3644 (PALIGNR128rr VR128:$src2, VR128:$src1,
3645 (SHUFFLE_get_palign_imm VR128:$src3))>,
3646 Requires<[HasSSSE3]>;
3647 def : Pat<(v16i8 (palign:$src3 VR128:$src1, VR128:$src2)),
3648 (PALIGNR128rr VR128:$src2, VR128:$src1,
3649 (SHUFFLE_get_palign_imm VR128:$src3))>,
3650 Requires<[HasSSSE3]>;
3653 //===---------------------------------------------------------------------===//
3654 // SSSE3 Misc Instructions
3655 //===---------------------------------------------------------------------===//
3657 // Thread synchronization
3658 let usesCustomInserter = 1 in {
3659 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
3660 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
3661 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
3662 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
3665 let Uses = [EAX, ECX, EDX] in
3666 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
3667 Requires<[HasSSE3]>;
3668 let Uses = [ECX, EAX] in
3669 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
3670 Requires<[HasSSE3]>;
3672 //===---------------------------------------------------------------------===//
3673 // Non-Instruction Patterns
3674 //===---------------------------------------------------------------------===//
3676 // extload f32 -> f64. This matches load+fextend because we have a hack in
3677 // the isel (PreprocessForFPConvert) that can introduce loads after dag
3679 // Since these loads aren't folded into the fextend, we have to match it
3681 let Predicates = [HasSSE2] in
3682 def : Pat<(fextend (loadf32 addr:$src)),
3683 (CVTSS2SDrm addr:$src)>;
3686 let Predicates = [HasXMMInt] in {
3687 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
3688 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
3689 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
3690 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
3691 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
3692 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
3693 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
3694 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
3695 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
3696 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
3697 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
3698 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
3699 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
3700 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
3701 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
3702 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
3703 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
3704 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
3705 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
3706 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
3707 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
3708 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
3709 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
3710 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
3711 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
3712 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
3713 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
3714 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
3715 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
3716 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
3719 let Predicates = [HasAVX] in {
3720 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
3723 // Move scalar to XMM zero-extended
3724 // movd to XMM register zero-extends
3725 let AddedComplexity = 15 in {
3726 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
3727 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
3728 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
3729 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
3730 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
3731 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
3732 (MOVSSrr (v4f32 (V_SET0PS)),
3733 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
3734 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
3735 (MOVSSrr (v4i32 (V_SET0PI)),
3736 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
3739 // Splat v2f64 / v2i64
3740 let AddedComplexity = 10 in {
3741 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
3742 (UNPCKLPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3743 def : Pat<(unpckh (v2f64 VR128:$src), (undef)),
3744 (UNPCKHPDrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3745 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
3746 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3747 def : Pat<(unpckh (v2i64 VR128:$src), (undef)),
3748 (PUNPCKHQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
3751 // Special unary SHUFPSrri case.
3752 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
3753 (SHUFPSrri VR128:$src1, VR128:$src1,
3754 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3755 let AddedComplexity = 5 in
3756 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3757 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3758 Requires<[HasSSE2]>;
3759 // Special unary SHUFPDrri case.
3760 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
3761 (SHUFPDrri VR128:$src1, VR128:$src1,
3762 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3763 Requires<[HasSSE2]>;
3764 // Special unary SHUFPDrri case.
3765 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
3766 (SHUFPDrri VR128:$src1, VR128:$src1,
3767 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3768 Requires<[HasSSE2]>;
3769 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3770 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3771 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3772 Requires<[HasSSE2]>;
3774 // Special binary v4i32 shuffle cases with SHUFPS.
3775 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
3776 (SHUFPSrri VR128:$src1, VR128:$src2,
3777 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3778 Requires<[HasSSE2]>;
3779 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (bc_v4i32 (memopv2i64 addr:$src2)))),
3780 (SHUFPSrmi VR128:$src1, addr:$src2,
3781 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3782 Requires<[HasSSE2]>;
3783 // Special binary v2i64 shuffle cases using SHUFPDrri.
3784 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
3785 (SHUFPDrri VR128:$src1, VR128:$src2,
3786 (SHUFFLE_get_shuf_imm VR128:$src3))>,
3787 Requires<[HasSSE2]>;
3789 // vector_shuffle v1, <undef>, <0, 0, 1, 1, ...>
3790 let AddedComplexity = 15 in {
3791 def : Pat<(v4i32 (unpckl_undef:$src2 VR128:$src, (undef))),
3792 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3793 Requires<[OptForSpeed, HasSSE2]>;
3794 def : Pat<(v4f32 (unpckl_undef:$src2 VR128:$src, (undef))),
3795 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3796 Requires<[OptForSpeed, HasSSE2]>;
3798 let AddedComplexity = 10 in {
3799 def : Pat<(v4f32 (unpckl_undef VR128:$src, (undef))),
3800 (UNPCKLPSrr VR128:$src, VR128:$src)>;
3801 def : Pat<(v16i8 (unpckl_undef VR128:$src, (undef))),
3802 (PUNPCKLBWrr VR128:$src, VR128:$src)>;
3803 def : Pat<(v8i16 (unpckl_undef VR128:$src, (undef))),
3804 (PUNPCKLWDrr VR128:$src, VR128:$src)>;
3805 def : Pat<(v4i32 (unpckl_undef VR128:$src, (undef))),
3806 (PUNPCKLDQrr VR128:$src, VR128:$src)>;
3809 // vector_shuffle v1, <undef>, <2, 2, 3, 3, ...>
3810 let AddedComplexity = 15 in {
3811 def : Pat<(v4i32 (unpckh_undef:$src2 VR128:$src, (undef))),
3812 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3813 Requires<[OptForSpeed, HasSSE2]>;
3814 def : Pat<(v4f32 (unpckh_undef:$src2 VR128:$src, (undef))),
3815 (PSHUFDri VR128:$src, (SHUFFLE_get_shuf_imm VR128:$src2))>,
3816 Requires<[OptForSpeed, HasSSE2]>;
3818 let AddedComplexity = 10 in {
3819 def : Pat<(v4f32 (unpckh_undef VR128:$src, (undef))),
3820 (UNPCKHPSrr VR128:$src, VR128:$src)>;
3821 def : Pat<(v16i8 (unpckh_undef VR128:$src, (undef))),
3822 (PUNPCKHBWrr VR128:$src, VR128:$src)>;
3823 def : Pat<(v8i16 (unpckh_undef VR128:$src, (undef))),
3824 (PUNPCKHWDrr VR128:$src, VR128:$src)>;
3825 def : Pat<(v4i32 (unpckh_undef VR128:$src, (undef))),
3826 (PUNPCKHDQrr VR128:$src, VR128:$src)>;
3829 let AddedComplexity = 20 in {
3830 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
3831 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
3832 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
3834 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
3835 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
3836 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
3838 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
3839 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
3840 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3841 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
3842 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
3845 let AddedComplexity = 20 in {
3846 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
3847 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
3848 (MOVLPSrm VR128:$src1, addr:$src2)>;
3849 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
3850 (MOVLPDrm VR128:$src1, addr:$src2)>;
3851 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
3852 (MOVLPSrm VR128:$src1, addr:$src2)>;
3853 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
3854 (MOVLPDrm VR128:$src1, addr:$src2)>;
3857 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
3858 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3859 (MOVLPSmr addr:$src1, VR128:$src2)>;
3860 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3861 (MOVLPDmr addr:$src1, VR128:$src2)>;
3862 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
3864 (MOVLPSmr addr:$src1, VR128:$src2)>;
3865 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
3866 (MOVLPDmr addr:$src1, VR128:$src2)>;
3868 let AddedComplexity = 15 in {
3869 // Setting the lowest element in the vector.
3870 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
3871 (MOVSSrr (v4i32 VR128:$src1),
3872 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
3873 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
3874 (MOVSDrr (v2i64 VR128:$src1),
3875 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
3877 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
3878 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
3879 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3880 Requires<[HasSSE2]>;
3881 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
3882 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
3883 Requires<[HasSSE2]>;
3886 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
3887 // fall back to this for SSE1)
3888 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
3889 (SHUFPSrri VR128:$src2, VR128:$src1,
3890 (SHUFFLE_get_shuf_imm VR128:$src3))>;
3892 // Set lowest element and zero upper elements.
3893 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
3894 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
3896 // vector -> vector casts
3897 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
3898 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
3899 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
3900 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
3902 // Use movaps / movups for SSE integer load / store (one byte shorter).
3903 let Predicates = [HasSSE1] in {
3904 def : Pat<(alignedloadv4i32 addr:$src),
3905 (MOVAPSrm addr:$src)>;
3906 def : Pat<(loadv4i32 addr:$src),
3907 (MOVUPSrm addr:$src)>;
3908 def : Pat<(alignedloadv2i64 addr:$src),
3909 (MOVAPSrm addr:$src)>;
3910 def : Pat<(loadv2i64 addr:$src),
3911 (MOVUPSrm addr:$src)>;
3913 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3914 (MOVAPSmr addr:$dst, VR128:$src)>;
3915 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3916 (MOVAPSmr addr:$dst, VR128:$src)>;
3917 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3918 (MOVAPSmr addr:$dst, VR128:$src)>;
3919 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3920 (MOVAPSmr addr:$dst, VR128:$src)>;
3921 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3922 (MOVUPSmr addr:$dst, VR128:$src)>;
3923 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3924 (MOVUPSmr addr:$dst, VR128:$src)>;
3925 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3926 (MOVUPSmr addr:$dst, VR128:$src)>;
3927 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3928 (MOVUPSmr addr:$dst, VR128:$src)>;
3931 // Use vmovaps/vmovups for AVX 128-bit integer load/store (one byte shorter).
3932 let Predicates = [HasAVX] in {
3933 def : Pat<(alignedloadv4i32 addr:$src),
3934 (VMOVAPSrm addr:$src)>;
3935 def : Pat<(loadv4i32 addr:$src),
3936 (VMOVUPSrm addr:$src)>;
3937 def : Pat<(alignedloadv2i64 addr:$src),
3938 (VMOVAPSrm addr:$src)>;
3939 def : Pat<(loadv2i64 addr:$src),
3940 (VMOVUPSrm addr:$src)>;
3942 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
3943 (VMOVAPSmr addr:$dst, VR128:$src)>;
3944 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
3945 (VMOVAPSmr addr:$dst, VR128:$src)>;
3946 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
3947 (VMOVAPSmr addr:$dst, VR128:$src)>;
3948 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
3949 (VMOVAPSmr addr:$dst, VR128:$src)>;
3950 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
3951 (VMOVUPSmr addr:$dst, VR128:$src)>;
3952 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
3953 (VMOVUPSmr addr:$dst, VR128:$src)>;
3954 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
3955 (VMOVUPSmr addr:$dst, VR128:$src)>;
3956 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
3957 (VMOVUPSmr addr:$dst, VR128:$src)>;
3960 //===----------------------------------------------------------------------===//
3961 // SSE4.1 - Packed Move with Sign/Zero Extend
3962 //===----------------------------------------------------------------------===//
3964 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
3965 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3966 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3967 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
3969 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3970 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3972 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
3976 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
3977 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
3979 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
3981 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
3983 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
3985 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
3987 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
3991 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
3992 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
3993 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
3994 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
3995 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
3996 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
3998 // Common patterns involving scalar load.
3999 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4000 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4001 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4002 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4004 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4005 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4006 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4007 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4009 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4010 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4011 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4012 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4014 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4015 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4016 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4017 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4019 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4020 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4021 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4022 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4024 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4025 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4026 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4027 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4030 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4031 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4032 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4033 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4035 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4036 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4038 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4042 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4043 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4045 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4047 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4049 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4053 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4054 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4055 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4056 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4058 // Common patterns involving scalar load
4059 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4060 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4061 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4062 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4064 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4065 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4066 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4067 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4070 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4071 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4072 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4073 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4075 // Expecting a i16 load any extended to i32 value.
4076 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4077 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4078 [(set VR128:$dst, (IntId (bitconvert
4079 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4083 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4084 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4086 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4089 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4090 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4092 // Common patterns involving scalar load
4093 def : Pat<(int_x86_sse41_pmovsxbq
4094 (bitconvert (v4i32 (X86vzmovl
4095 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4096 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4098 def : Pat<(int_x86_sse41_pmovzxbq
4099 (bitconvert (v4i32 (X86vzmovl
4100 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4101 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4103 //===----------------------------------------------------------------------===//
4104 // SSE4.1 - Extract Instructions
4105 //===----------------------------------------------------------------------===//
4107 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4108 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4109 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4110 (ins VR128:$src1, i32i8imm:$src2),
4111 !strconcat(OpcodeStr,
4112 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4113 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4115 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4116 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4117 !strconcat(OpcodeStr,
4118 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4121 // There's an AssertZext in the way of writing the store pattern
4122 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4125 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4126 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4127 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4128 (ins VR128:$src1, i32i8imm:$src2),
4129 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4132 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4135 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4136 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4137 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4138 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4139 !strconcat(OpcodeStr,
4140 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4143 // There's an AssertZext in the way of writing the store pattern
4144 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4147 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4148 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4150 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4153 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4154 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4155 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4156 (ins VR128:$src1, i32i8imm:$src2),
4157 !strconcat(OpcodeStr,
4158 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4160 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4161 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4162 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4163 !strconcat(OpcodeStr,
4164 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4165 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4166 addr:$dst)]>, OpSize;
4169 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4170 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4172 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4174 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4175 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4176 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4177 (ins VR128:$src1, i32i8imm:$src2),
4178 !strconcat(OpcodeStr,
4179 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4181 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4182 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4183 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4184 !strconcat(OpcodeStr,
4185 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4186 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4187 addr:$dst)]>, OpSize, REX_W;
4190 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4191 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4193 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4195 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4197 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4198 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4199 (ins VR128:$src1, i32i8imm:$src2),
4200 !strconcat(OpcodeStr,
4201 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4203 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4205 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4206 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4207 !strconcat(OpcodeStr,
4208 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4209 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4210 addr:$dst)]>, OpSize;
4213 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4214 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4215 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4216 (ins VR128:$src1, i32i8imm:$src2),
4217 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4220 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4222 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4223 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4226 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4227 Requires<[HasSSE41]>;
4229 //===----------------------------------------------------------------------===//
4230 // SSE4.1 - Insert Instructions
4231 //===----------------------------------------------------------------------===//
4233 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4234 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4235 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4237 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4239 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4241 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4242 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4243 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4245 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4247 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4249 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4250 imm:$src3))]>, OpSize;
4253 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4254 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4255 let Constraints = "$src1 = $dst" in
4256 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4258 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4259 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4260 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4262 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4264 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4266 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4268 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4269 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4271 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4273 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4275 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4276 imm:$src3)))]>, OpSize;
4279 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4280 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4281 let Constraints = "$src1 = $dst" in
4282 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4284 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4285 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4286 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4288 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4290 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4292 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4294 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4295 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4297 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4299 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4301 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4302 imm:$src3)))]>, OpSize;
4305 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4306 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4307 let Constraints = "$src1 = $dst" in
4308 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4310 // insertps has a few different modes, there's the first two here below which
4311 // are optimized inserts that won't zero arbitrary elements in the destination
4312 // vector. The next one matches the intrinsic and could zero arbitrary elements
4313 // in the target vector.
4314 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4315 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4316 (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4318 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4320 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4322 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4324 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4325 (ins VR128:$src1, f32mem:$src2, i32i8imm:$src3),
4327 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4329 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4331 (X86insrtps VR128:$src1,
4332 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4333 imm:$src3))]>, OpSize;
4336 let Constraints = "$src1 = $dst" in
4337 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4338 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4339 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4341 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4342 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4344 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4345 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4346 Requires<[HasSSE41]>;
4348 //===----------------------------------------------------------------------===//
4349 // SSE4.1 - Round Instructions
4350 //===----------------------------------------------------------------------===//
4352 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4353 X86MemOperand x86memop, RegisterClass RC,
4354 PatFrag mem_frag32, PatFrag mem_frag64,
4355 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4356 // Intrinsic operation, reg.
4357 // Vector intrinsic operation, reg
4358 def PSr : SS4AIi8<opcps, MRMSrcReg,
4359 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4360 !strconcat(OpcodeStr,
4361 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4362 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4365 // Vector intrinsic operation, mem
4366 def PSm : Ii8<opcps, MRMSrcMem,
4367 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4368 !strconcat(OpcodeStr,
4369 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4371 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4373 Requires<[HasSSE41]>;
4375 // Vector intrinsic operation, reg
4376 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4377 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4378 !strconcat(OpcodeStr,
4379 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4380 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4383 // Vector intrinsic operation, mem
4384 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4385 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4386 !strconcat(OpcodeStr,
4387 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4389 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4393 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4394 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4395 // Intrinsic operation, reg.
4396 // Vector intrinsic operation, reg
4397 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4398 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4399 !strconcat(OpcodeStr,
4400 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4403 // Vector intrinsic operation, mem
4404 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4405 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4406 !strconcat(OpcodeStr,
4407 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4408 []>, TA, OpSize, Requires<[HasSSE41]>;
4410 // Vector intrinsic operation, reg
4411 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
4412 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4413 !strconcat(OpcodeStr,
4414 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4417 // Vector intrinsic operation, mem
4418 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
4419 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4420 !strconcat(OpcodeStr,
4421 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4425 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
4428 Intrinsic F64Int, bit Is2Addr = 1> {
4429 // Intrinsic operation, reg.
4430 def SSr : SS4AIi8<opcss, MRMSrcReg,
4431 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4433 !strconcat(OpcodeStr,
4434 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4435 !strconcat(OpcodeStr,
4436 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4437 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4440 // Intrinsic operation, mem.
4441 def SSm : SS4AIi8<opcss, MRMSrcMem,
4442 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4444 !strconcat(OpcodeStr,
4445 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4446 !strconcat(OpcodeStr,
4447 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4449 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
4452 // Intrinsic operation, reg.
4453 def SDr : SS4AIi8<opcsd, MRMSrcReg,
4454 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4456 !strconcat(OpcodeStr,
4457 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4458 !strconcat(OpcodeStr,
4459 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4460 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
4463 // Intrinsic operation, mem.
4464 def SDm : SS4AIi8<opcsd, MRMSrcMem,
4465 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4467 !strconcat(OpcodeStr,
4468 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4469 !strconcat(OpcodeStr,
4470 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4472 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
4476 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
4478 // Intrinsic operation, reg.
4479 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
4480 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4481 !strconcat(OpcodeStr,
4482 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4485 // Intrinsic operation, mem.
4486 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
4487 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
4488 !strconcat(OpcodeStr,
4489 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4492 // Intrinsic operation, reg.
4493 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
4494 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
4495 !strconcat(OpcodeStr,
4496 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4499 // Intrinsic operation, mem.
4500 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
4501 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
4502 !strconcat(OpcodeStr,
4503 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4507 // FP round - roundss, roundps, roundsd, roundpd
4508 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4510 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
4511 memopv4f32, memopv2f64,
4512 int_x86_sse41_round_ps,
4513 int_x86_sse41_round_pd>, VEX;
4514 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
4515 memopv8f32, memopv4f64,
4516 int_x86_avx_round_ps_256,
4517 int_x86_avx_round_pd_256>, VEX;
4518 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
4519 int_x86_sse41_round_ss,
4520 int_x86_sse41_round_sd, 0>, VEX_4V;
4522 // Instructions for the assembler
4523 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
4525 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
4527 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
4530 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
4531 memopv4f32, memopv2f64,
4532 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
4533 let Constraints = "$src1 = $dst" in
4534 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
4535 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
4537 //===----------------------------------------------------------------------===//
4538 // SSE4.1 - Packed Bit Test
4539 //===----------------------------------------------------------------------===//
4541 // ptest instruction we'll lower to this in X86ISelLowering primarily from
4542 // the intel intrinsic that corresponds to this.
4543 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4544 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4545 "vptest\t{$src2, $src1|$src1, $src2}",
4546 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4548 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4549 "vptest\t{$src2, $src1|$src1, $src2}",
4550 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4553 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
4554 "vptest\t{$src2, $src1|$src1, $src2}",
4555 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
4557 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
4558 "vptest\t{$src2, $src1|$src1, $src2}",
4559 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
4563 let Defs = [EFLAGS] in {
4564 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
4565 "ptest \t{$src2, $src1|$src1, $src2}",
4566 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
4568 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
4569 "ptest \t{$src2, $src1|$src1, $src2}",
4570 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
4574 // The bit test instructions below are AVX only
4575 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
4576 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
4577 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
4578 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4579 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
4580 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
4581 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
4582 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
4586 let Defs = [EFLAGS], isAsmParserOnly = 1, Predicates = [HasAVX] in {
4587 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
4588 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
4589 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
4590 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
4593 //===----------------------------------------------------------------------===//
4594 // SSE4.1 - Misc Instructions
4595 //===----------------------------------------------------------------------===//
4597 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
4598 "popcnt{w}\t{$src, $dst|$dst, $src}",
4599 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
4600 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
4601 "popcnt{w}\t{$src, $dst|$dst, $src}",
4602 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
4604 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
4605 "popcnt{l}\t{$src, $dst|$dst, $src}",
4606 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
4607 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
4608 "popcnt{l}\t{$src, $dst|$dst, $src}",
4609 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
4611 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
4612 "popcnt{q}\t{$src, $dst|$dst, $src}",
4613 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
4614 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
4615 "popcnt{q}\t{$src, $dst|$dst, $src}",
4616 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
4620 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
4621 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
4622 Intrinsic IntId128> {
4623 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4625 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4626 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
4627 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4629 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4632 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
4635 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4636 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
4637 int_x86_sse41_phminposuw>, VEX;
4638 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
4639 int_x86_sse41_phminposuw>;
4641 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
4642 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
4643 Intrinsic IntId128, bit Is2Addr = 1> {
4644 let isCommutable = 1 in
4645 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4646 (ins VR128:$src1, VR128:$src2),
4648 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4649 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4650 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
4651 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4652 (ins VR128:$src1, i128mem:$src2),
4654 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4655 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4657 (IntId128 VR128:$src1,
4658 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4661 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4662 let isCommutable = 0 in
4663 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
4665 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
4667 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
4669 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
4671 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
4673 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
4675 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
4677 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
4679 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
4681 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
4683 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
4687 let Constraints = "$src1 = $dst" in {
4688 let isCommutable = 0 in
4689 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
4690 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
4691 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
4692 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
4693 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
4694 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
4695 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
4696 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
4697 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
4698 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
4699 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
4702 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
4703 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
4704 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
4705 (PCMPEQQrm VR128:$src1, addr:$src2)>;
4707 /// SS48I_binop_rm - Simple SSE41 binary operator.
4708 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
4709 ValueType OpVT, bit Is2Addr = 1> {
4710 let isCommutable = 1 in
4711 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4712 (ins VR128:$src1, VR128:$src2),
4714 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4715 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4716 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
4718 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4719 (ins VR128:$src1, i128mem:$src2),
4721 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4722 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4723 [(set VR128:$dst, (OpNode VR128:$src1,
4724 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
4728 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4729 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
4730 let Constraints = "$src1 = $dst" in
4731 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
4733 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
4734 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
4735 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
4736 X86MemOperand x86memop, bit Is2Addr = 1> {
4737 let isCommutable = 1 in
4738 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
4739 (ins RC:$src1, RC:$src2, i32i8imm:$src3),
4741 !strconcat(OpcodeStr,
4742 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4743 !strconcat(OpcodeStr,
4744 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4745 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
4747 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
4748 (ins RC:$src1, x86memop:$src2, i32i8imm:$src3),
4750 !strconcat(OpcodeStr,
4751 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4752 !strconcat(OpcodeStr,
4753 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4756 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
4760 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4761 let isCommutable = 0 in {
4762 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
4763 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4764 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
4765 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4766 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
4767 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4768 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
4769 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
4770 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
4771 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4772 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
4773 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4775 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
4776 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4777 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
4778 VR128, memopv16i8, i128mem, 0>, VEX_4V;
4779 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
4780 VR256, memopv32i8, i256mem, 0>, VEX_4V;
4783 let Constraints = "$src1 = $dst" in {
4784 let isCommutable = 0 in {
4785 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
4786 VR128, memopv16i8, i128mem>;
4787 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
4788 VR128, memopv16i8, i128mem>;
4789 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
4790 VR128, memopv16i8, i128mem>;
4791 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
4792 VR128, memopv16i8, i128mem>;
4794 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
4795 VR128, memopv16i8, i128mem>;
4796 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
4797 VR128, memopv16i8, i128mem>;
4800 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
4801 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4802 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
4803 RegisterClass RC, X86MemOperand x86memop,
4804 PatFrag mem_frag, Intrinsic IntId> {
4805 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
4806 (ins RC:$src1, RC:$src2, RC:$src3),
4807 !strconcat(OpcodeStr,
4808 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4809 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
4810 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4812 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
4813 (ins RC:$src1, x86memop:$src2, RC:$src3),
4814 !strconcat(OpcodeStr,
4815 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4817 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
4819 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
4823 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
4824 memopv16i8, int_x86_sse41_blendvpd>;
4825 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
4826 memopv16i8, int_x86_sse41_blendvps>;
4827 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
4828 memopv16i8, int_x86_sse41_pblendvb>;
4829 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
4830 memopv32i8, int_x86_avx_blendv_pd_256>;
4831 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
4832 memopv32i8, int_x86_avx_blendv_ps_256>;
4834 /// SS41I_ternary_int - SSE 4.1 ternary operator
4835 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
4836 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4837 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
4838 (ins VR128:$src1, VR128:$src2),
4839 !strconcat(OpcodeStr,
4840 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4841 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
4844 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
4845 (ins VR128:$src1, i128mem:$src2),
4846 !strconcat(OpcodeStr,
4847 "\t{%xmm0, $src2, $dst|$dst, $src2, %xmm0}"),
4850 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
4854 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
4855 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
4856 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
4858 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4859 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4860 "vmovntdqa\t{$src, $dst|$dst, $src}",
4861 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4863 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4864 "movntdqa\t{$src, $dst|$dst, $src}",
4865 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
4868 //===----------------------------------------------------------------------===//
4869 // SSE4.2 - Compare Instructions
4870 //===----------------------------------------------------------------------===//
4872 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
4873 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
4874 Intrinsic IntId128, bit Is2Addr = 1> {
4875 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
4876 (ins VR128:$src1, VR128:$src2),
4878 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4879 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4880 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4882 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
4883 (ins VR128:$src1, i128mem:$src2),
4885 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4886 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4888 (IntId128 VR128:$src1,
4889 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4892 let isAsmParserOnly = 1, Predicates = [HasAVX] in
4893 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
4895 let Constraints = "$src1 = $dst" in
4896 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
4898 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
4899 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
4900 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
4901 (PCMPGTQrm VR128:$src1, addr:$src2)>;
4903 //===----------------------------------------------------------------------===//
4904 // SSE4.2 - String/text Processing Instructions
4905 //===----------------------------------------------------------------------===//
4907 // Packed Compare Implicit Length Strings, Return Mask
4908 multiclass pseudo_pcmpistrm<string asm> {
4909 def REG : PseudoI<(outs VR128:$dst),
4910 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4911 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
4913 def MEM : PseudoI<(outs VR128:$dst),
4914 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4915 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
4916 VR128:$src1, (load addr:$src2), imm:$src3))]>;
4919 let Defs = [EFLAGS], usesCustomInserter = 1 in {
4920 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
4921 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
4924 let Defs = [XMM0, EFLAGS], isAsmParserOnly = 1,
4925 Predicates = [HasAVX] in {
4926 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4927 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4928 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4929 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4930 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4931 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
4934 let Defs = [XMM0, EFLAGS] in {
4935 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
4936 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4937 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4938 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
4939 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4940 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
4943 // Packed Compare Explicit Length Strings, Return Mask
4944 multiclass pseudo_pcmpestrm<string asm> {
4945 def REG : PseudoI<(outs VR128:$dst),
4946 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4947 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4948 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
4949 def MEM : PseudoI<(outs VR128:$dst),
4950 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4951 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
4952 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
4955 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
4956 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
4957 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
4960 let isAsmParserOnly = 1, Predicates = [HasAVX],
4961 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4962 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4963 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4964 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4965 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4966 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4967 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
4970 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
4971 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
4972 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
4973 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4974 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
4975 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
4976 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
4979 // Packed Compare Implicit Length Strings, Return Index
4980 let Defs = [ECX, EFLAGS] in {
4981 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
4982 def rr : SS42AI<0x63, MRMSrcReg, (outs),
4983 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4984 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4985 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
4986 (implicit EFLAGS)]>, OpSize;
4987 def rm : SS42AI<0x63, MRMSrcMem, (outs),
4988 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4989 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
4990 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
4991 (implicit EFLAGS)]>, OpSize;
4995 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
4996 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
4998 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5000 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5002 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5004 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5006 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5010 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5011 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5012 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5013 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5014 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5015 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5017 // Packed Compare Explicit Length Strings, Return Index
5018 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5019 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5020 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5021 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5022 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5023 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5024 (implicit EFLAGS)]>, OpSize;
5025 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5026 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5027 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5029 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5030 (implicit EFLAGS)]>, OpSize;
5034 let isAsmParserOnly = 1, Predicates = [HasAVX] in {
5035 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5037 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5039 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5041 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5043 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5045 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5049 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5050 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5051 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5052 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5053 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5054 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5056 //===----------------------------------------------------------------------===//
5057 // SSE4.2 - CRC Instructions
5058 //===----------------------------------------------------------------------===//
5060 // No CRC instructions have AVX equivalents
5062 // crc intrinsic instruction
5063 // This set of instructions are only rm, the only difference is the size
5065 let Constraints = "$src1 = $dst" in {
5066 def CRC32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5067 (ins GR32:$src1, i8mem:$src2),
5068 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5070 (int_x86_sse42_crc32_8 GR32:$src1,
5071 (load addr:$src2)))]>;
5072 def CRC32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5073 (ins GR32:$src1, GR8:$src2),
5074 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5076 (int_x86_sse42_crc32_8 GR32:$src1, GR8:$src2))]>;
5077 def CRC32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5078 (ins GR32:$src1, i16mem:$src2),
5079 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5081 (int_x86_sse42_crc32_16 GR32:$src1,
5082 (load addr:$src2)))]>,
5084 def CRC32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5085 (ins GR32:$src1, GR16:$src2),
5086 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5088 (int_x86_sse42_crc32_16 GR32:$src1, GR16:$src2))]>,
5090 def CRC32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5091 (ins GR32:$src1, i32mem:$src2),
5092 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5094 (int_x86_sse42_crc32_32 GR32:$src1,
5095 (load addr:$src2)))]>;
5096 def CRC32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5097 (ins GR32:$src1, GR32:$src2),
5098 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5100 (int_x86_sse42_crc32_32 GR32:$src1, GR32:$src2))]>;
5101 def CRC64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5102 (ins GR64:$src1, i8mem:$src2),
5103 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5105 (int_x86_sse42_crc64_8 GR64:$src1,
5106 (load addr:$src2)))]>,
5108 def CRC64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5109 (ins GR64:$src1, GR8:$src2),
5110 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5112 (int_x86_sse42_crc64_8 GR64:$src1, GR8:$src2))]>,
5114 def CRC64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5115 (ins GR64:$src1, i64mem:$src2),
5116 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5118 (int_x86_sse42_crc64_64 GR64:$src1,
5119 (load addr:$src2)))]>,
5121 def CRC64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5122 (ins GR64:$src1, GR64:$src2),
5123 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5125 (int_x86_sse42_crc64_64 GR64:$src1, GR64:$src2))]>,
5129 //===----------------------------------------------------------------------===//
5130 // AES-NI Instructions
5131 //===----------------------------------------------------------------------===//
5133 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5134 Intrinsic IntId128, bit Is2Addr = 1> {
5135 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5136 (ins VR128:$src1, VR128:$src2),
5138 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5139 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5140 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5142 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5143 (ins VR128:$src1, i128mem:$src2),
5145 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5146 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5148 (IntId128 VR128:$src1,
5149 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5152 // Perform One Round of an AES Encryption/Decryption Flow
5153 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5154 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5155 int_x86_aesni_aesenc, 0>, VEX_4V;
5156 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5157 int_x86_aesni_aesenclast, 0>, VEX_4V;
5158 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5159 int_x86_aesni_aesdec, 0>, VEX_4V;
5160 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5161 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5164 let Constraints = "$src1 = $dst" in {
5165 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5166 int_x86_aesni_aesenc>;
5167 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5168 int_x86_aesni_aesenclast>;
5169 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5170 int_x86_aesni_aesdec>;
5171 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5172 int_x86_aesni_aesdeclast>;
5175 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5176 (AESENCrr VR128:$src1, VR128:$src2)>;
5177 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5178 (AESENCrm VR128:$src1, addr:$src2)>;
5179 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5180 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5181 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5182 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5183 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5184 (AESDECrr VR128:$src1, VR128:$src2)>;
5185 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5186 (AESDECrm VR128:$src1, addr:$src2)>;
5187 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5188 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5189 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5190 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5192 // Perform the AES InvMixColumn Transformation
5193 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5194 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5196 "vaesimc\t{$src1, $dst|$dst, $src1}",
5198 (int_x86_aesni_aesimc VR128:$src1))]>,
5200 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5201 (ins i128mem:$src1),
5202 "vaesimc\t{$src1, $dst|$dst, $src1}",
5204 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5207 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5209 "aesimc\t{$src1, $dst|$dst, $src1}",
5211 (int_x86_aesni_aesimc VR128:$src1))]>,
5213 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5214 (ins i128mem:$src1),
5215 "aesimc\t{$src1, $dst|$dst, $src1}",
5217 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5220 // AES Round Key Generation Assist
5221 let isAsmParserOnly = 1, Predicates = [HasAVX, HasAES] in {
5222 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5223 (ins VR128:$src1, i8imm:$src2),
5224 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5226 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5228 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5229 (ins i128mem:$src1, i8imm:$src2),
5230 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5232 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5236 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5237 (ins VR128:$src1, i8imm:$src2),
5238 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5240 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5242 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5243 (ins i128mem:$src1, i8imm:$src2),
5244 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5246 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5250 //===----------------------------------------------------------------------===//
5251 // CLMUL Instructions
5252 //===----------------------------------------------------------------------===//
5254 // Only the AVX version of CLMUL instructions are described here.
5256 // Carry-less Multiplication instructions
5257 let isAsmParserOnly = 1 in {
5258 def VPCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5259 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5260 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5263 def VPCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5264 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5265 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5269 multiclass avx_vpclmul<string asm> {
5270 def rr : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
5271 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5274 def rm : I<0, Pseudo, (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
5275 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5278 defm VPCLMULHQHQDQ : avx_vpclmul<"vpclmulhqhqdq">;
5279 defm VPCLMULHQLQDQ : avx_vpclmul<"vpclmulhqlqdq">;
5280 defm VPCLMULLQHQDQ : avx_vpclmul<"vpclmullqhqdq">;
5281 defm VPCLMULLQLQDQ : avx_vpclmul<"vpclmullqlqdq">;
5283 } // isAsmParserOnly
5285 //===----------------------------------------------------------------------===//
5287 //===----------------------------------------------------------------------===//
5289 let isAsmParserOnly = 1 in {
5291 // Load from memory and broadcast to all elements of the destination operand
5292 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5293 X86MemOperand x86memop, Intrinsic Int> :
5294 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5295 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5296 [(set RC:$dst, (Int addr:$src))]>, VEX;
5298 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5299 int_x86_avx_vbroadcastss>;
5300 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5301 int_x86_avx_vbroadcastss_256>;
5302 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5303 int_x86_avx_vbroadcast_sd_256>;
5304 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5305 int_x86_avx_vbroadcastf128_pd_256>;
5307 // Insert packed floating-point values
5308 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5309 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5310 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5312 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5313 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5314 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5317 // Extract packed floating-point values
5318 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5319 (ins VR256:$src1, i8imm:$src2),
5320 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5322 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5323 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5324 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5327 // Conditional SIMD Packed Loads and Stores
5328 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
5329 Intrinsic IntLd, Intrinsic IntLd256,
5330 Intrinsic IntSt, Intrinsic IntSt256,
5331 PatFrag pf128, PatFrag pf256> {
5332 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
5333 (ins VR128:$src1, f128mem:$src2),
5334 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5335 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
5337 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
5338 (ins VR256:$src1, f256mem:$src2),
5339 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5340 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
5342 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
5343 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
5344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5345 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
5346 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
5347 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
5348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5349 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
5352 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
5353 int_x86_avx_maskload_ps,
5354 int_x86_avx_maskload_ps_256,
5355 int_x86_avx_maskstore_ps,
5356 int_x86_avx_maskstore_ps_256,
5357 memopv4f32, memopv8f32>;
5358 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
5359 int_x86_avx_maskload_pd,
5360 int_x86_avx_maskload_pd_256,
5361 int_x86_avx_maskstore_pd,
5362 int_x86_avx_maskstore_pd_256,
5363 memopv2f64, memopv4f64>;
5365 // Permute Floating-Point Values
5366 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
5367 RegisterClass RC, X86MemOperand x86memop_f,
5368 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
5369 Intrinsic IntVar, Intrinsic IntImm> {
5370 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
5371 (ins RC:$src1, RC:$src2),
5372 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5373 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
5374 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
5375 (ins RC:$src1, x86memop_i:$src2),
5376 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5377 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
5379 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
5380 (ins RC:$src1, i8imm:$src2),
5381 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5382 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
5383 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
5384 (ins x86memop_f:$src1, i8imm:$src2),
5385 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5386 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
5389 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
5390 memopv4f32, memopv4i32,
5391 int_x86_avx_vpermilvar_ps,
5392 int_x86_avx_vpermil_ps>;
5393 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
5394 memopv8f32, memopv8i32,
5395 int_x86_avx_vpermilvar_ps_256,
5396 int_x86_avx_vpermil_ps_256>;
5397 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
5398 memopv2f64, memopv2i64,
5399 int_x86_avx_vpermilvar_pd,
5400 int_x86_avx_vpermil_pd>;
5401 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
5402 memopv4f64, memopv4i64,
5403 int_x86_avx_vpermilvar_pd_256,
5404 int_x86_avx_vpermil_pd_256>;
5406 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
5407 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5408 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5410 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
5411 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
5412 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5415 // Zero All YMM registers
5416 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
5417 [(int_x86_avx_vzeroall)]>, VEX, VEX_L, Requires<[HasAVX]>;
5419 // Zero Upper bits of YMM registers
5420 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
5421 [(int_x86_avx_vzeroupper)]>, VEX, Requires<[HasAVX]>;
5423 } // isAsmParserOnly
5425 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5426 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5427 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5428 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5429 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5430 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5432 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
5433 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5434 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
5435 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5436 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
5437 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
5439 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5440 (VBROADCASTF128 addr:$src)>;
5442 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
5443 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5444 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
5445 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5446 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
5447 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
5449 def : Pat<(int_x86_avx_vperm2f128_ps_256
5450 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
5451 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5452 def : Pat<(int_x86_avx_vperm2f128_pd_256
5453 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
5454 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5455 def : Pat<(int_x86_avx_vperm2f128_si_256
5456 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
5457 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
5459 //===----------------------------------------------------------------------===//
5460 // SSE Shuffle pattern fragments
5461 //===----------------------------------------------------------------------===//
5463 // This is part of a "work in progress" refactoring. The idea is that all
5464 // vector shuffles are going to be translated into target specific nodes and
5465 // directly matched by the patterns below (which can be changed along the way)
5466 // The AVX version of some but not all of them are described here, and more
5467 // should come in a near future.
5469 // Shuffle with PSHUFD instruction folding loads. The first two patterns match
5470 // SSE2 loads, which are always promoted to v2i64. The last one should match
5471 // the SSE1 case, where the only legal load is v4f32, but there is no PSHUFD
5472 // in SSE2, how does it ever worked? Anyway, the pattern will remain here until
5473 // we investigate further.
5474 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5476 (VPSHUFDmi addr:$src1, imm:$imm)>, Requires<[HasAVX]>;
5477 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
5479 (PSHUFDmi addr:$src1, imm:$imm)>;
5480 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
5482 (PSHUFDmi addr:$src1, imm:$imm)>; // FIXME: has this ever worked?
5484 // Shuffle with PSHUFD instruction.
5485 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5486 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5487 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5488 (PSHUFDri VR128:$src1, imm:$imm)>;
5490 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5491 (VPSHUFDri VR128:$src1, imm:$imm)>, Requires<[HasAVX]>;
5492 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
5493 (PSHUFDri VR128:$src1, imm:$imm)>;
5495 // Shuffle with SHUFPD instruction.
5496 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5497 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5498 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5499 def : Pat<(v2f64 (X86Shufps VR128:$src1,
5500 (memopv2f64 addr:$src2), (i8 imm:$imm))),
5501 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
5503 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5504 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5505 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5506 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5508 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5509 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5510 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5511 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
5513 // Shuffle with SHUFPS instruction.
5514 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5515 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5516 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5517 def : Pat<(v4f32 (X86Shufps VR128:$src1,
5518 (memopv4f32 addr:$src2), (i8 imm:$imm))),
5519 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5521 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5522 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5523 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5524 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5526 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5527 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5528 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>, Requires<[HasAVX]>;
5529 def : Pat<(v4i32 (X86Shufps VR128:$src1,
5530 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
5531 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
5533 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5534 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>, Requires<[HasAVX]>;
5535 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5536 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
5538 // Shuffle with MOVHLPS instruction
5539 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
5540 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5541 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
5542 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
5544 // Shuffle with MOVDDUP instruction
5545 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5546 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5547 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5548 (MOVDDUPrm addr:$src)>;
5550 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5551 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5552 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5553 (MOVDDUPrm addr:$src)>;
5555 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5556 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5557 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5558 (MOVDDUPrm addr:$src)>;
5560 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5561 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5562 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5563 (MOVDDUPrm addr:$src)>;
5565 def : Pat<(X86Movddup (bc_v2f64
5566 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5567 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5568 def : Pat<(X86Movddup (bc_v2f64
5569 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5570 (MOVDDUPrm addr:$src)>;
5573 // Shuffle with UNPCKLPS
5574 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5575 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5576 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
5577 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5579 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5580 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5581 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
5582 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
5584 // Shuffle with UNPCKHPS
5585 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5586 (VUNPCKHPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5587 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
5588 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
5590 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5591 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5592 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
5593 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
5595 // Shuffle with UNPCKLPD
5596 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5597 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5598 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
5599 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5601 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5602 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5603 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
5604 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
5606 // Shuffle with UNPCKHPD
5607 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5608 (VUNPCKLPSrm VR128:$src1, addr:$src2)>, Requires<[HasAVX]>;
5609 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
5610 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
5612 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5613 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>, Requires<[HasAVX]>;
5614 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
5615 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
5617 // Shuffle with PUNPCKLBW
5618 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1,
5619 (bc_v16i8 (memopv2i64 addr:$src2)))),
5620 (PUNPCKLBWrm VR128:$src1, addr:$src2)>;
5621 def : Pat<(v16i8 (X86Punpcklbw VR128:$src1, VR128:$src2)),
5622 (PUNPCKLBWrr VR128:$src1, VR128:$src2)>;
5624 // Shuffle with PUNPCKLWD
5625 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1,
5626 (bc_v8i16 (memopv2i64 addr:$src2)))),
5627 (PUNPCKLWDrm VR128:$src1, addr:$src2)>;
5628 def : Pat<(v8i16 (X86Punpcklwd VR128:$src1, VR128:$src2)),
5629 (PUNPCKLWDrr VR128:$src1, VR128:$src2)>;
5631 // Shuffle with PUNPCKLDQ
5632 def : Pat<(v4i32 (X86Punpckldq VR128:$src1,
5633 (bc_v4i32 (memopv2i64 addr:$src2)))),
5634 (PUNPCKLDQrm VR128:$src1, addr:$src2)>;
5635 def : Pat<(v4i32 (X86Punpckldq VR128:$src1, VR128:$src2)),
5636 (PUNPCKLDQrr VR128:$src1, VR128:$src2)>;
5638 // Shuffle with PUNPCKLQDQ
5639 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, (memopv2i64 addr:$src2))),
5640 (PUNPCKLQDQrm VR128:$src1, addr:$src2)>;
5641 def : Pat<(v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)),
5642 (PUNPCKLQDQrr VR128:$src1, VR128:$src2)>;
5644 // Shuffle with PUNPCKHBW
5645 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1,
5646 (bc_v16i8 (memopv2i64 addr:$src2)))),
5647 (PUNPCKHBWrm VR128:$src1, addr:$src2)>;
5648 def : Pat<(v16i8 (X86Punpckhbw VR128:$src1, VR128:$src2)),
5649 (PUNPCKHBWrr VR128:$src1, VR128:$src2)>;
5651 // Shuffle with PUNPCKHWD
5652 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1,
5653 (bc_v8i16 (memopv2i64 addr:$src2)))),
5654 (PUNPCKHWDrm VR128:$src1, addr:$src2)>;
5655 def : Pat<(v8i16 (X86Punpckhwd VR128:$src1, VR128:$src2)),
5656 (PUNPCKHWDrr VR128:$src1, VR128:$src2)>;
5658 // Shuffle with PUNPCKHDQ
5659 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1,
5660 (bc_v4i32 (memopv2i64 addr:$src2)))),
5661 (PUNPCKHDQrm VR128:$src1, addr:$src2)>;
5662 def : Pat<(v4i32 (X86Punpckhdq VR128:$src1, VR128:$src2)),
5663 (PUNPCKHDQrr VR128:$src1, VR128:$src2)>;
5665 // Shuffle with PUNPCKHQDQ
5666 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, (memopv2i64 addr:$src2))),
5667 (PUNPCKHQDQrm VR128:$src1, addr:$src2)>;
5668 def : Pat<(v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)),
5669 (PUNPCKHQDQrr VR128:$src1, VR128:$src2)>;
5671 // Shuffle with MOVLHPS
5672 def : Pat<(X86Movlhps VR128:$src1,
5673 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5674 (MOVHPSrm VR128:$src1, addr:$src2)>;
5675 def : Pat<(X86Movlhps VR128:$src1,
5676 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
5677 (MOVHPSrm VR128:$src1, addr:$src2)>;
5678 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
5679 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5680 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
5681 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
5682 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
5683 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
5685 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the problem
5686 // is during lowering, where it's not possible to recognize the load fold cause
5687 // it has two uses through a bitcast. One use disappears at isel time and the
5688 // fold opportunity reappears.
5689 def : Pat<(v2f64 (X86Movddup VR128:$src)),
5690 (UNPCKLPDrr VR128:$src, VR128:$src)>;
5692 // Shuffle with MOVLHPD
5693 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
5694 (scalar_to_vector (loadf64 addr:$src2)))),
5695 (MOVHPDrm VR128:$src1, addr:$src2)>;
5697 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
5698 // is during lowering, where it's not possible to recognize the load fold cause
5699 // it has two uses through a bitcast. One use disappears at isel time and the
5700 // fold opportunity reappears.
5701 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
5702 (scalar_to_vector (loadf64 addr:$src2)))),
5703 (MOVHPDrm VR128:$src1, addr:$src2)>;
5705 // Shuffle with MOVSS
5706 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
5707 (MOVSSrr VR128:$src1, FR32:$src2)>;
5708 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
5709 (MOVSSrr (v4i32 VR128:$src1),
5710 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
5711 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
5712 (MOVSSrr (v4f32 VR128:$src1),
5713 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
5714 // FIXME: Instead of a X86Movss there should be a X86Movlps here, the problem
5715 // is during lowering, where it's not possible to recognize the load fold cause
5716 // it has two uses through a bitcast. One use disappears at isel time and the
5717 // fold opportunity reappears.
5718 def : Pat<(X86Movss VR128:$src1,
5719 (bc_v4i32 (v2i64 (load addr:$src2)))),
5720 (MOVLPSrm VR128:$src1, addr:$src2)>;
5722 // Shuffle with MOVSD
5723 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
5724 (MOVSDrr VR128:$src1, FR64:$src2)>;
5725 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
5726 (MOVSDrr (v2i64 VR128:$src1),
5727 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
5728 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
5729 (MOVSDrr (v2f64 VR128:$src1),
5730 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
5731 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
5732 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5733 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
5734 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
5736 // Shuffle with MOVSHDUP
5737 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
5738 (MOVSHDUPrr VR128:$src)>;
5739 def : Pat<(X86Movshdup (bc_v4i32 (memopv2i64 addr:$src))),
5740 (MOVSHDUPrm addr:$src)>;
5742 def : Pat<(v4f32 (X86Movshdup VR128:$src)),
5743 (MOVSHDUPrr VR128:$src)>;
5744 def : Pat<(X86Movshdup (memopv4f32 addr:$src)),
5745 (MOVSHDUPrm addr:$src)>;
5747 // Shuffle with MOVSLDUP
5748 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
5749 (MOVSLDUPrr VR128:$src)>;
5750 def : Pat<(X86Movsldup (bc_v4i32 (memopv2i64 addr:$src))),
5751 (MOVSLDUPrm addr:$src)>;
5753 def : Pat<(v4f32 (X86Movsldup VR128:$src)),
5754 (MOVSLDUPrr VR128:$src)>;
5755 def : Pat<(X86Movsldup (memopv4f32 addr:$src)),
5756 (MOVSLDUPrm addr:$src)>;
5758 // Shuffle with PSHUFHW
5759 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
5760 (PSHUFHWri VR128:$src, imm:$imm)>;
5761 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5762 (PSHUFHWmi addr:$src, imm:$imm)>;
5764 // Shuffle with PSHUFLW
5765 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
5766 (PSHUFLWri VR128:$src, imm:$imm)>;
5767 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)), (i8 imm:$imm))),
5768 (PSHUFLWmi addr:$src, imm:$imm)>;
5770 // Shuffle with PALIGN
5771 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5772 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5773 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5774 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5775 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5776 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5777 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5778 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5780 // Shuffle with MOVLPS
5781 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
5782 (MOVLPSrm VR128:$src1, addr:$src2)>;
5783 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
5784 (MOVLPSrm VR128:$src1, addr:$src2)>;
5785 def : Pat<(X86Movlps VR128:$src1,
5786 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
5787 (MOVLPSrm VR128:$src1, addr:$src2)>;
5788 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
5789 // is during lowering, where it's not possible to recognize the load fold cause
5790 // it has two uses through a bitcast. One use disappears at isel time and the
5791 // fold opportunity reappears.
5792 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
5793 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
5795 // Shuffle with MOVLPD
5796 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5797 (MOVLPDrm VR128:$src1, addr:$src2)>;
5798 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
5799 (MOVLPDrm VR128:$src1, addr:$src2)>;
5800 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
5801 (scalar_to_vector (loadf64 addr:$src2)))),
5802 (MOVLPDrm VR128:$src1, addr:$src2)>;
5804 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
5805 def : Pat<(store (f64 (vector_extract
5806 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5807 (MOVHPSmr addr:$dst, VR128:$src)>;
5808 def : Pat<(store (f64 (vector_extract
5809 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
5810 (MOVHPDmr addr:$dst, VR128:$src)>;
5812 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
5813 (MOVLPSmr addr:$src1, VR128:$src2)>;
5814 def : Pat<(store (v4i32 (X86Movlps
5815 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
5816 (MOVLPSmr addr:$src1, VR128:$src2)>;
5818 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5819 (MOVLPDmr addr:$src1, VR128:$src2)>;
5820 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
5821 (MOVLPDmr addr:$src1, VR128:$src2)>;