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 // Non-instruction patterns
120 //===----------------------------------------------------------------------===//
122 // A vector extract of the first f32 position is a subregister copy
123 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
124 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 // A 128-bit subvector extract from the first 256-bit vector position
127 // is a subregister copy that needs no instruction.
128 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
129 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
130 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
131 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
133 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
134 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
135 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
136 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
138 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
139 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
140 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
141 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
143 // A 128-bit subvector insert to the first 256-bit vector position
144 // is a subregister copy that needs no instruction.
145 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
146 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
147 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
148 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
149 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
150 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
151 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
152 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
153 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
154 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
155 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
156 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 // Implicitly promote a 32-bit scalar to a vector.
159 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
160 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
161 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
162 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
163 // Implicitly promote a 64-bit scalar to a vector.
164 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
165 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
166 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
167 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 // Bitcasts between 128-bit vector types. Return the original type since
170 // no instruction is needed for the conversion
171 let Predicates = [HasXMMInt] in {
172 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
173 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
174 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
175 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
178 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
179 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
183 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
184 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
188 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
189 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
193 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
194 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
198 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
199 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
204 // Bitcasts between 256-bit vector types. Return the original type since
205 // no instruction is needed for the conversion
206 let Predicates = [HasAVX] in {
207 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
208 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
209 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
210 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
213 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
214 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
218 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
219 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
223 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
224 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
228 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
229 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
233 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
234 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
239 //===----------------------------------------------------------------------===//
240 // AVX & SSE - Zero/One Vectors
241 //===----------------------------------------------------------------------===//
243 // Alias instructions that map zero vector to pxor / xorp* for sse.
244 // We set canFoldAsLoad because this can be converted to a constant-pool
245 // load of an all-zeros value if folding it would be beneficial.
246 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
247 // JIT implementation, it does not expand the instructions below like
248 // X86MCInstLower does.
249 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
250 isCodeGenOnly = 1 in {
251 def V_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
252 [(set VR128:$dst, (v4f32 immAllZerosV))]>;
253 def V_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
254 [(set VR128:$dst, (v2f64 immAllZerosV))]>;
255 let ExeDomain = SSEPackedInt in
256 def V_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
257 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
260 // The same as done above but for AVX. The 128-bit versions are the
261 // same, but re-encoded. The 256-bit does not support PI version, and
262 // doesn't need it because on sandy bridge the register is set to zero
263 // at the rename stage without using any execution unit, so SET0PSY
264 // and SET0PDY can be used for vector int instructions without penalty
265 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
266 // JIT implementatioan, it does not expand the instructions below like
267 // X86MCInstLower does.
268 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
269 isCodeGenOnly = 1, Predicates = [HasAVX] in {
270 def AVX_SET0PS : PSI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
271 [(set VR128:$dst, (v4f32 immAllZerosV))]>, VEX_4V;
272 def AVX_SET0PD : PDI<0x57, MRMInitReg, (outs VR128:$dst), (ins), "",
273 [(set VR128:$dst, (v2f64 immAllZerosV))]>, VEX_4V;
274 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
275 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
276 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
277 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
278 let ExeDomain = SSEPackedInt in
279 def AVX_SET0PI : PDI<0xEF, MRMInitReg, (outs VR128:$dst), (ins), "",
280 [(set VR128:$dst, (v4i32 immAllZerosV))]>;
283 def : Pat<(v2i64 immAllZerosV), (V_SET0PI)>;
284 def : Pat<(v8i16 immAllZerosV), (V_SET0PI)>;
285 def : Pat<(v16i8 immAllZerosV), (V_SET0PI)>;
287 // AVX has no support for 256-bit integer instructions, but since the 128-bit
288 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
289 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
290 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
291 (SUBREG_TO_REG (i32 0), (AVX_SET0PI), sub_xmm)>;
293 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
294 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
295 (SUBREG_TO_REG (i64 0), (AVX_SET0PI), sub_xmm)>;
297 //===----------------------------------------------------------------------===//
298 // SSE 1 & 2 - Move Instructions
299 //===----------------------------------------------------------------------===//
301 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
302 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
303 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
305 // Loading from memory automatically zeroing upper bits.
306 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
307 PatFrag mem_pat, string OpcodeStr> :
308 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
309 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
310 [(set RC:$dst, (mem_pat addr:$src))]>;
312 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
313 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
314 // is used instead. Register-to-register movss/movsd is not modeled as an
315 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
316 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
317 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
318 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
319 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
320 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
322 let canFoldAsLoad = 1, isReMaterializable = 1 in {
323 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
325 let AddedComplexity = 20 in
326 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
329 let Constraints = "$src1 = $dst" in {
330 def MOVSSrr : sse12_move_rr<FR32, v4f32,
331 "movss\t{$src2, $dst|$dst, $src2}">, XS;
332 def MOVSDrr : sse12_move_rr<FR64, v2f64,
333 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
336 let canFoldAsLoad = 1, isReMaterializable = 1 in {
337 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
339 let AddedComplexity = 20 in
340 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
343 let AddedComplexity = 15 in {
344 // Extract the low 32-bit value from one vector and insert it into another.
345 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
346 (MOVSSrr (v4f32 VR128:$src1),
347 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
348 // Extract the low 64-bit value from one vector and insert it into another.
349 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
350 (MOVSDrr (v2f64 VR128:$src1),
351 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
354 let AddedComplexity = 20 in {
355 let Predicates = [HasSSE1] in {
356 // MOVSSrm zeros the high parts of the register; represent this
357 // with SUBREG_TO_REG.
358 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
359 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
360 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
361 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
362 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
363 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
365 let Predicates = [HasSSE2] in {
366 // MOVSDrm zeros the high parts of the register; represent this
367 // with SUBREG_TO_REG.
368 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
369 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
370 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
371 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
372 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
373 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
374 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
375 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
376 def : Pat<(v2f64 (X86vzload addr:$src)),
377 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
381 let AddedComplexity = 20, Predicates = [HasAVX] in {
382 // MOVSSrm zeros the high parts of the register; represent this
383 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
384 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
385 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
386 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
387 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
388 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
389 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
390 // MOVSDrm zeros the high parts of the register; represent this
391 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
392 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
393 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
394 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
395 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
396 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
397 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
398 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
399 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
400 def : Pat<(v2f64 (X86vzload addr:$src)),
401 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
402 // Represent the same patterns above but in the form they appear for
404 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
405 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
406 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
407 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
408 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
409 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
412 // Store scalar value to memory.
413 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
414 "movss\t{$src, $dst|$dst, $src}",
415 [(store FR32:$src, addr:$dst)]>;
416 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
417 "movsd\t{$src, $dst|$dst, $src}",
418 [(store FR64:$src, addr:$dst)]>;
420 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
421 "movss\t{$src, $dst|$dst, $src}",
422 [(store FR32:$src, addr:$dst)]>, XS, VEX;
423 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
424 "movsd\t{$src, $dst|$dst, $src}",
425 [(store FR64:$src, addr:$dst)]>, XD, VEX;
427 // Extract and store.
428 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
431 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
432 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
435 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
437 // Move Aligned/Unaligned floating point values
438 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
439 X86MemOperand x86memop, PatFrag ld_frag,
440 string asm, Domain d,
441 bit IsReMaterializable = 1> {
442 let neverHasSideEffects = 1 in
443 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
444 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
445 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
446 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
447 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
448 [(set RC:$dst, (ld_frag addr:$src))], d>;
451 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
452 "movaps", SSEPackedSingle>, TB, VEX;
453 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
454 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
455 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
456 "movups", SSEPackedSingle>, TB, VEX;
457 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
458 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
460 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
461 "movaps", SSEPackedSingle>, TB, VEX;
462 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
463 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
464 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
465 "movups", SSEPackedSingle>, TB, VEX;
466 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
467 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
468 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
469 "movaps", SSEPackedSingle>, TB;
470 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
471 "movapd", SSEPackedDouble>, TB, OpSize;
472 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
473 "movups", SSEPackedSingle>, TB;
474 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
475 "movupd", SSEPackedDouble, 0>, TB, OpSize;
477 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
478 "movaps\t{$src, $dst|$dst, $src}",
479 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
480 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
481 "movapd\t{$src, $dst|$dst, $src}",
482 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
483 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
484 "movups\t{$src, $dst|$dst, $src}",
485 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
486 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
487 "movupd\t{$src, $dst|$dst, $src}",
488 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
489 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
490 "movaps\t{$src, $dst|$dst, $src}",
491 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
492 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
493 "movapd\t{$src, $dst|$dst, $src}",
494 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
495 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
496 "movups\t{$src, $dst|$dst, $src}",
497 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
498 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
499 "movupd\t{$src, $dst|$dst, $src}",
500 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
502 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
503 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
504 (VMOVUPSYmr addr:$dst, VR256:$src)>;
506 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
507 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
508 (VMOVUPDYmr addr:$dst, VR256:$src)>;
510 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
511 "movaps\t{$src, $dst|$dst, $src}",
512 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
513 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
514 "movapd\t{$src, $dst|$dst, $src}",
515 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
516 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
517 "movups\t{$src, $dst|$dst, $src}",
518 [(store (v4f32 VR128:$src), addr:$dst)]>;
519 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
520 "movupd\t{$src, $dst|$dst, $src}",
521 [(store (v2f64 VR128:$src), addr:$dst)]>;
523 // Intrinsic forms of MOVUPS/D load and store
524 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
525 (ins f128mem:$dst, VR128:$src),
526 "movups\t{$src, $dst|$dst, $src}",
527 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
528 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
529 (ins f128mem:$dst, VR128:$src),
530 "movupd\t{$src, $dst|$dst, $src}",
531 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
533 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
534 "movups\t{$src, $dst|$dst, $src}",
535 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
536 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
537 "movupd\t{$src, $dst|$dst, $src}",
538 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
540 // Move Low/High packed floating point values
541 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
542 PatFrag mov_frag, string base_opc,
544 def PSrm : PI<opc, MRMSrcMem,
545 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
546 !strconcat(base_opc, "s", asm_opr),
549 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
550 SSEPackedSingle>, TB;
552 def PDrm : PI<opc, MRMSrcMem,
553 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
554 !strconcat(base_opc, "d", asm_opr),
555 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
556 (scalar_to_vector (loadf64 addr:$src2)))))],
557 SSEPackedDouble>, TB, OpSize;
560 let AddedComplexity = 20 in {
561 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
562 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
563 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
564 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
566 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
567 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
568 "\t{$src2, $dst|$dst, $src2}">;
569 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
570 "\t{$src2, $dst|$dst, $src2}">;
573 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
574 "movlps\t{$src, $dst|$dst, $src}",
575 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
576 (iPTR 0))), addr:$dst)]>, VEX;
577 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
578 "movlpd\t{$src, $dst|$dst, $src}",
579 [(store (f64 (vector_extract (v2f64 VR128:$src),
580 (iPTR 0))), addr:$dst)]>, VEX;
581 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
582 "movlps\t{$src, $dst|$dst, $src}",
583 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
584 (iPTR 0))), addr:$dst)]>;
585 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
586 "movlpd\t{$src, $dst|$dst, $src}",
587 [(store (f64 (vector_extract (v2f64 VR128:$src),
588 (iPTR 0))), addr:$dst)]>;
590 // v2f64 extract element 1 is always custom lowered to unpack high to low
591 // and extract element 0 so the non-store version isn't too horrible.
592 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
593 "movhps\t{$src, $dst|$dst, $src}",
594 [(store (f64 (vector_extract
595 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
596 (undef)), (iPTR 0))), addr:$dst)]>,
598 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
599 "movhpd\t{$src, $dst|$dst, $src}",
600 [(store (f64 (vector_extract
601 (v2f64 (unpckh VR128:$src, (undef))),
602 (iPTR 0))), addr:$dst)]>,
604 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
605 "movhps\t{$src, $dst|$dst, $src}",
606 [(store (f64 (vector_extract
607 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
608 (undef)), (iPTR 0))), addr:$dst)]>;
609 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
610 "movhpd\t{$src, $dst|$dst, $src}",
611 [(store (f64 (vector_extract
612 (v2f64 (unpckh VR128:$src, (undef))),
613 (iPTR 0))), addr:$dst)]>;
615 let AddedComplexity = 20 in {
616 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
617 (ins VR128:$src1, VR128:$src2),
618 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
620 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
622 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
623 (ins VR128:$src1, VR128:$src2),
624 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
626 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
629 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
630 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
631 (ins VR128:$src1, VR128:$src2),
632 "movlhps\t{$src2, $dst|$dst, $src2}",
634 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
635 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
636 (ins VR128:$src1, VR128:$src2),
637 "movhlps\t{$src2, $dst|$dst, $src2}",
639 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
642 let Predicates = [HasAVX] in {
644 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
645 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
646 def : Pat<(X86Movlhps VR128:$src1,
647 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
648 (VMOVHPSrm VR128:$src1, addr:$src2)>;
649 def : Pat<(X86Movlhps VR128:$src1,
650 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
651 (VMOVHPSrm VR128:$src1, addr:$src2)>;
654 let AddedComplexity = 20 in {
655 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
656 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
657 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
658 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
660 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
661 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
662 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
664 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
665 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
666 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
667 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
668 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
669 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
672 let AddedComplexity = 20 in {
673 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
674 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
675 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
677 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
678 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
679 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
680 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
681 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
684 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
685 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
686 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
687 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
690 let Predicates = [HasSSE1] in {
692 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
693 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
695 def : Pat<(X86Movlhps VR128:$src1,
696 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
697 (MOVHPSrm VR128:$src1, addr:$src2)>;
698 def : Pat<(X86Movlhps VR128:$src1,
699 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
700 (MOVHPSrm VR128:$src1, addr:$src2)>;
703 let AddedComplexity = 20 in {
704 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
705 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
706 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
707 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
709 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
710 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
711 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
713 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
714 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
715 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
716 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
717 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
718 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
721 let AddedComplexity = 20 in {
722 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
723 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
724 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
726 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
727 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
728 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
729 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
730 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
733 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
734 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
735 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
736 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
739 //===----------------------------------------------------------------------===//
740 // SSE 1 & 2 - Conversion Instructions
741 //===----------------------------------------------------------------------===//
743 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
744 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
746 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
747 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
748 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
749 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
752 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
753 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
754 string asm, Domain d> {
755 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
756 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
757 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
758 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
761 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
762 X86MemOperand x86memop, string asm> {
763 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
764 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
765 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
766 (ins DstRC:$src1, x86memop:$src),
767 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
770 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
771 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
772 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
773 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
775 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
776 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
777 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
778 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
781 // The assembler can recognize rr 64-bit instructions by seeing a rxx
782 // register, but the same isn't true when only using memory operands,
783 // provide other assembly "l" and "q" forms to address this explicitly
784 // where appropriate to do so.
785 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
787 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
789 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
791 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
793 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
796 let Predicates = [HasAVX] in {
797 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
798 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
799 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
800 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
801 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
802 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
803 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
804 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
806 def : Pat<(f32 (sint_to_fp GR32:$src)),
807 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
808 def : Pat<(f32 (sint_to_fp GR64:$src)),
809 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
810 def : Pat<(f64 (sint_to_fp GR32:$src)),
811 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
812 def : Pat<(f64 (sint_to_fp GR64:$src)),
813 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
816 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
817 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
818 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
819 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
820 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
821 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
822 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
823 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
824 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
825 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
826 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
827 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
828 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
829 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
830 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
831 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
833 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
834 // and/or XMM operand(s).
836 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
837 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
839 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
840 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
841 [(set DstRC:$dst, (Int SrcRC:$src))]>;
842 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
843 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
844 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
847 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
848 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
849 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
850 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
852 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
853 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
854 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
855 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
856 (ins DstRC:$src1, x86memop:$src2),
858 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
859 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
860 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
863 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
864 f128mem, load, "cvtsd2si">, XD, VEX;
865 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
866 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
869 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
870 // Get rid of this hack or rename the intrinsics, there are several
871 // intructions that only match with the intrinsic form, why create duplicates
872 // to let them be recognized by the assembler?
873 let Pattern = []<dag> in {
874 defm VCVTSD2SI : sse12_cvt_s<0x2D, FR64, GR32, undef, f64mem, load,
875 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
876 defm VCVTSD2SI64 : sse12_cvt_s<0x2D, FR64, GR64, undef, f64mem, load,
877 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
879 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
880 f128mem, load, "cvtsd2si{l}">, XD;
881 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
882 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
885 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
886 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
887 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
888 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
890 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
891 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
892 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
893 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
896 let Constraints = "$src1 = $dst" in {
897 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
898 int_x86_sse_cvtsi2ss, i32mem, loadi32,
900 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
901 int_x86_sse_cvtsi642ss, i64mem, loadi64,
902 "cvtsi2ss{q}">, XS, REX_W;
903 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
904 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
906 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
907 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
908 "cvtsi2sd">, XD, REX_W;
913 // Aliases for intrinsics
914 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
915 f32mem, load, "cvttss2si">, XS, VEX;
916 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
917 int_x86_sse_cvttss2si64, f32mem, load,
918 "cvttss2si">, XS, VEX, VEX_W;
919 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
920 f128mem, load, "cvttsd2si">, XD, VEX;
921 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
922 int_x86_sse2_cvttsd2si64, f128mem, load,
923 "cvttsd2si">, XD, VEX, VEX_W;
924 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
925 f32mem, load, "cvttss2si">, XS;
926 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
927 int_x86_sse_cvttss2si64, f32mem, load,
928 "cvttss2si{q}">, XS, REX_W;
929 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
930 f128mem, load, "cvttsd2si">, XD;
931 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
932 int_x86_sse2_cvttsd2si64, f128mem, load,
933 "cvttsd2si{q}">, XD, REX_W;
935 let Pattern = []<dag> in {
936 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
937 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
938 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
939 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
941 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
942 "cvtdq2ps\t{$src, $dst|$dst, $src}",
943 SSEPackedSingle>, TB, VEX;
944 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
945 "cvtdq2ps\t{$src, $dst|$dst, $src}",
946 SSEPackedSingle>, TB, VEX;
949 let Pattern = []<dag> in {
950 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
951 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
952 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
953 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
954 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
955 "cvtdq2ps\t{$src, $dst|$dst, $src}",
956 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
959 let Predicates = [HasSSE1] in {
960 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
961 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
962 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
963 (CVTSS2SIrm addr:$src)>;
964 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
965 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
966 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
967 (CVTSS2SI64rm addr:$src)>;
970 let Predicates = [HasAVX] in {
971 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
972 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
973 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
974 (VCVTSS2SIrm addr:$src)>;
975 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
976 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
977 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
978 (VCVTSS2SI64rm addr:$src)>;
983 // Convert scalar double to scalar single
984 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
985 (ins FR64:$src1, FR64:$src2),
986 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
988 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
989 (ins FR64:$src1, f64mem:$src2),
990 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
991 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
992 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
995 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
996 "cvtsd2ss\t{$src, $dst|$dst, $src}",
997 [(set FR32:$dst, (fround FR64:$src))]>;
998 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
999 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1000 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1001 Requires<[HasSSE2, OptForSize]>;
1003 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1004 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1006 let Constraints = "$src1 = $dst" in
1007 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1008 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1010 // Convert scalar single to scalar double
1011 // SSE2 instructions with XS prefix
1012 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1013 (ins FR32:$src1, FR32:$src2),
1014 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1015 []>, XS, Requires<[HasAVX]>, VEX_4V;
1016 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1017 (ins FR32:$src1, f32mem:$src2),
1018 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1019 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
1021 let Predicates = [HasAVX] in {
1022 def : Pat<(f64 (fextend FR32:$src)),
1023 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1024 def : Pat<(fextend (loadf32 addr:$src)),
1025 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1026 def : Pat<(extloadf32 addr:$src),
1027 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1030 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1031 "cvtss2sd\t{$src, $dst|$dst, $src}",
1032 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1033 Requires<[HasSSE2]>;
1034 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1035 "cvtss2sd\t{$src, $dst|$dst, $src}",
1036 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1037 Requires<[HasSSE2, OptForSize]>;
1039 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1040 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1041 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1042 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1043 VR128:$src2))]>, XS, VEX_4V,
1045 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1046 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1047 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1048 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1049 (load addr:$src2)))]>, XS, VEX_4V,
1051 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1052 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1053 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1054 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1055 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1056 VR128:$src2))]>, XS,
1057 Requires<[HasSSE2]>;
1058 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1059 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1060 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1061 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1062 (load addr:$src2)))]>, XS,
1063 Requires<[HasSSE2]>;
1066 def : Pat<(extloadf32 addr:$src),
1067 (CVTSS2SDrr (MOVSSrm addr:$src))>,
1068 Requires<[HasSSE2, OptForSpeed]>;
1070 // Convert doubleword to packed single/double fp
1071 // SSE2 instructions without OpSize prefix
1072 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1073 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1074 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1075 TB, VEX, Requires<[HasAVX]>;
1076 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1077 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1078 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1079 (bitconvert (memopv2i64 addr:$src))))]>,
1080 TB, VEX, Requires<[HasAVX]>;
1081 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1082 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1083 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1084 TB, Requires<[HasSSE2]>;
1085 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1086 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1087 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1088 (bitconvert (memopv2i64 addr:$src))))]>,
1089 TB, Requires<[HasSSE2]>;
1091 // FIXME: why the non-intrinsic version is described as SSE3?
1092 // SSE2 instructions with XS prefix
1093 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1094 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1095 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1096 XS, VEX, Requires<[HasAVX]>;
1097 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1098 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1099 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1100 (bitconvert (memopv2i64 addr:$src))))]>,
1101 XS, VEX, Requires<[HasAVX]>;
1102 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1103 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1104 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1105 XS, Requires<[HasSSE2]>;
1106 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1107 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1108 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1109 (bitconvert (memopv2i64 addr:$src))))]>,
1110 XS, Requires<[HasSSE2]>;
1113 // Convert packed single/double fp to doubleword
1114 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1115 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1116 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1117 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1118 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1119 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1120 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1121 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1122 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1123 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1124 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1125 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1127 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1128 "cvtps2dq\t{$src, $dst|$dst, $src}",
1129 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1131 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1133 "cvtps2dq\t{$src, $dst|$dst, $src}",
1134 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1135 (memop addr:$src)))]>, VEX;
1136 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1137 "cvtps2dq\t{$src, $dst|$dst, $src}",
1138 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1139 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1140 "cvtps2dq\t{$src, $dst|$dst, $src}",
1141 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1142 (memop addr:$src)))]>;
1144 // SSE2 packed instructions with XD prefix
1145 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1146 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1147 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1148 XD, VEX, Requires<[HasAVX]>;
1149 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1150 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1151 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1152 (memop addr:$src)))]>,
1153 XD, VEX, Requires<[HasAVX]>;
1154 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1155 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1156 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1157 XD, Requires<[HasSSE2]>;
1158 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1159 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1160 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1161 (memop addr:$src)))]>,
1162 XD, Requires<[HasSSE2]>;
1165 // Convert with truncation packed single/double fp to doubleword
1166 // SSE2 packed instructions with XS prefix
1167 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1168 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1169 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1170 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1171 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1172 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1173 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1174 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1175 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1176 "cvttps2dq\t{$src, $dst|$dst, $src}",
1178 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1179 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1180 "cvttps2dq\t{$src, $dst|$dst, $src}",
1182 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1184 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1185 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1187 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1188 XS, VEX, Requires<[HasAVX]>;
1189 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1190 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1191 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1192 (memop addr:$src)))]>,
1193 XS, VEX, Requires<[HasAVX]>;
1195 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1196 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
1197 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1198 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
1200 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1201 (Int_VCVTDQ2PSrr VR128:$src)>, Requires<[HasAVX]>;
1202 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1203 (VCVTTPS2DQrr VR128:$src)>, Requires<[HasAVX]>;
1204 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1205 (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
1206 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1207 (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
1209 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
1211 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1212 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1214 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1216 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1217 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1218 (memop addr:$src)))]>, VEX;
1219 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1220 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1221 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1222 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1223 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1224 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1225 (memop addr:$src)))]>;
1227 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1228 // register, but the same isn't true when using memory operands instead.
1229 // Provide other assembly rr and rm forms to address this explicitly.
1230 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1231 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1232 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1233 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1236 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1237 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1238 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1239 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1242 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1243 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1244 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1245 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1247 // Convert packed single to packed double
1248 let Predicates = [HasAVX] in {
1249 // SSE2 instructions without OpSize prefix
1250 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1251 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1252 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1253 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1254 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1255 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1256 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1257 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1259 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1260 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1261 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1262 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1264 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1265 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1266 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1267 TB, VEX, Requires<[HasAVX]>;
1268 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1269 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1270 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1271 (load addr:$src)))]>,
1272 TB, VEX, Requires<[HasAVX]>;
1273 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1274 "cvtps2pd\t{$src, $dst|$dst, $src}",
1275 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1276 TB, Requires<[HasSSE2]>;
1277 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1278 "cvtps2pd\t{$src, $dst|$dst, $src}",
1279 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1280 (load addr:$src)))]>,
1281 TB, Requires<[HasSSE2]>;
1283 // Convert packed double to packed single
1284 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1285 // register, but the same isn't true when using memory operands instead.
1286 // Provide other assembly rr and rm forms to address this explicitly.
1287 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1288 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1289 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1290 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1293 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1294 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1295 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1296 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1299 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1300 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1301 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1302 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1303 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1304 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1305 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1306 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1309 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1310 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1311 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1312 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1314 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1315 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1316 (memop addr:$src)))]>;
1317 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1318 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1319 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1320 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1321 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1322 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1323 (memop addr:$src)))]>;
1325 // AVX 256-bit register conversion intrinsics
1326 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1327 // whenever possible to avoid declaring two versions of each one.
1328 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1329 (VCVTDQ2PSYrr VR256:$src)>;
1330 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1331 (VCVTDQ2PSYrm addr:$src)>;
1333 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1334 (VCVTPD2PSYrr VR256:$src)>;
1335 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1336 (VCVTPD2PSYrm addr:$src)>;
1338 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1339 (VCVTPS2DQYrr VR256:$src)>;
1340 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1341 (VCVTPS2DQYrm addr:$src)>;
1343 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1344 (VCVTPS2PDYrr VR128:$src)>;
1345 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1346 (VCVTPS2PDYrm addr:$src)>;
1348 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1349 (VCVTTPD2DQYrr VR256:$src)>;
1350 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1351 (VCVTTPD2DQYrm addr:$src)>;
1353 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1354 (VCVTTPS2DQYrr VR256:$src)>;
1355 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1356 (VCVTTPS2DQYrm addr:$src)>;
1358 // Match fround and fextend for 128/256-bit conversions
1359 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1360 (VCVTPD2PSYrr VR256:$src)>;
1361 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1362 (VCVTPD2PSYrm addr:$src)>;
1364 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1365 (VCVTPS2PDYrr VR128:$src)>;
1366 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1367 (VCVTPS2PDYrm addr:$src)>;
1369 //===----------------------------------------------------------------------===//
1370 // SSE 1 & 2 - Compare Instructions
1371 //===----------------------------------------------------------------------===//
1373 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1374 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1375 string asm, string asm_alt> {
1376 let isAsmParserOnly = 1 in {
1377 def rr : SIi8<0xC2, MRMSrcReg,
1378 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1381 def rm : SIi8<0xC2, MRMSrcMem,
1382 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1386 // Accept explicit immediate argument form instead of comparison code.
1387 def rr_alt : SIi8<0xC2, MRMSrcReg,
1388 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1391 def rm_alt : SIi8<0xC2, MRMSrcMem,
1392 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1396 let neverHasSideEffects = 1 in {
1397 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1398 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1399 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1401 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1402 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1403 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1407 let Constraints = "$src1 = $dst" in {
1408 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1409 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1410 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1411 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1412 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1413 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1414 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1415 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1416 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1417 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1418 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1419 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1420 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1421 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1422 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1423 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1425 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1426 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1427 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1428 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1429 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1430 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1431 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1432 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1433 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1434 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1435 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1436 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1437 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1440 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1441 Intrinsic Int, string asm> {
1442 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1443 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1444 [(set VR128:$dst, (Int VR128:$src1,
1445 VR128:$src, imm:$cc))]>;
1446 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1447 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1448 [(set VR128:$dst, (Int VR128:$src1,
1449 (load addr:$src), imm:$cc))]>;
1452 // Aliases to match intrinsics which expect XMM operand(s).
1453 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1454 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1456 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1457 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1459 let Constraints = "$src1 = $dst" in {
1460 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1461 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1462 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1463 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1467 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1468 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1469 ValueType vt, X86MemOperand x86memop,
1470 PatFrag ld_frag, string OpcodeStr, Domain d> {
1471 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1472 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1473 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1474 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1475 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1476 [(set EFLAGS, (OpNode (vt RC:$src1),
1477 (ld_frag addr:$src2)))], d>;
1480 let Defs = [EFLAGS] in {
1481 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1482 "ucomiss", SSEPackedSingle>, TB, VEX;
1483 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1484 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1485 let Pattern = []<dag> in {
1486 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1487 "comiss", SSEPackedSingle>, TB, VEX;
1488 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1489 "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1492 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1493 load, "ucomiss", SSEPackedSingle>, TB, VEX;
1494 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1495 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1497 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1498 load, "comiss", SSEPackedSingle>, TB, VEX;
1499 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1500 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1501 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1502 "ucomiss", SSEPackedSingle>, TB;
1503 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1504 "ucomisd", SSEPackedDouble>, TB, OpSize;
1506 let Pattern = []<dag> in {
1507 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1508 "comiss", SSEPackedSingle>, TB;
1509 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1510 "comisd", SSEPackedDouble>, TB, OpSize;
1513 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1514 load, "ucomiss", SSEPackedSingle>, TB;
1515 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1516 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1518 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1519 "comiss", SSEPackedSingle>, TB;
1520 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1521 "comisd", SSEPackedDouble>, TB, OpSize;
1522 } // Defs = [EFLAGS]
1524 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1525 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1526 Intrinsic Int, string asm, string asm_alt,
1528 let isAsmParserOnly = 1 in {
1529 def rri : PIi8<0xC2, MRMSrcReg,
1530 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1531 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1532 def rmi : PIi8<0xC2, MRMSrcMem,
1533 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1534 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1537 // Accept explicit immediate argument form instead of comparison code.
1538 def rri_alt : PIi8<0xC2, MRMSrcReg,
1539 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1541 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1542 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1546 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1547 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1548 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1549 SSEPackedSingle>, TB, VEX_4V;
1550 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1551 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1552 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1553 SSEPackedDouble>, TB, OpSize, VEX_4V;
1554 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1555 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1556 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1557 SSEPackedSingle>, TB, VEX_4V;
1558 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1559 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1560 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1561 SSEPackedDouble>, TB, OpSize, VEX_4V;
1562 let Constraints = "$src1 = $dst" in {
1563 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1564 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1565 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1566 SSEPackedSingle>, TB;
1567 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1568 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1569 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1570 SSEPackedDouble>, TB, OpSize;
1573 let Predicates = [HasSSE1] in {
1574 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1575 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1576 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1577 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1580 let Predicates = [HasSSE2] in {
1581 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1582 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1583 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1584 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1587 let Predicates = [HasAVX] in {
1588 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1589 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1590 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1591 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1592 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1593 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1594 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1595 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1597 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
1598 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
1599 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
1600 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
1601 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
1602 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
1603 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
1604 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
1607 //===----------------------------------------------------------------------===//
1608 // SSE 1 & 2 - Shuffle Instructions
1609 //===----------------------------------------------------------------------===//
1611 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1612 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1613 ValueType vt, string asm, PatFrag mem_frag,
1614 Domain d, bit IsConvertibleToThreeAddress = 0> {
1615 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1616 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1617 [(set RC:$dst, (vt (shufp:$src3
1618 RC:$src1, (mem_frag addr:$src2))))], d>;
1619 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1620 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1621 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1623 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1626 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1627 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1628 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1629 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1630 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1631 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1632 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1633 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1634 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1635 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1636 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1637 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1639 let Constraints = "$src1 = $dst" in {
1640 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1641 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1642 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1644 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1645 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1646 memopv2f64, SSEPackedDouble>, TB, OpSize;
1649 let Predicates = [HasSSE1] in {
1650 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1651 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1652 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1653 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1654 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1655 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1656 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1657 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1658 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1659 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1660 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1661 // fall back to this for SSE1)
1662 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1663 (SHUFPSrri VR128:$src2, VR128:$src1,
1664 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1665 // Special unary SHUFPSrri case.
1666 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1667 (SHUFPSrri VR128:$src1, VR128:$src1,
1668 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1671 let Predicates = [HasSSE2] in {
1672 // Special binary v4i32 shuffle cases with SHUFPS.
1673 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1674 (SHUFPSrri VR128:$src1, VR128:$src2,
1675 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1676 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1677 (bc_v4i32 (memopv2i64 addr:$src2)))),
1678 (SHUFPSrmi VR128:$src1, addr:$src2,
1679 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1680 // Special unary SHUFPDrri cases.
1681 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1682 (SHUFPDrri VR128:$src1, VR128:$src1,
1683 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1684 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1685 (SHUFPDrri VR128:$src1, VR128:$src1,
1686 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1687 // Special binary v2i64 shuffle cases using SHUFPDrri.
1688 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1689 (SHUFPDrri VR128:$src1, VR128:$src2,
1690 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1691 // Generic SHUFPD patterns
1692 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1693 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1694 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1695 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1696 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1697 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1698 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1701 let Predicates = [HasAVX] in {
1702 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1703 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1704 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1705 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1706 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1707 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1708 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1709 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1710 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1711 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1712 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1713 // fall back to this for SSE1)
1714 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1715 (VSHUFPSrri VR128:$src2, VR128:$src1,
1716 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1717 // Special unary SHUFPSrri case.
1718 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1719 (VSHUFPSrri VR128:$src1, VR128:$src1,
1720 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1721 // Special binary v4i32 shuffle cases with SHUFPS.
1722 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1723 (VSHUFPSrri VR128:$src1, VR128:$src2,
1724 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1725 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1726 (bc_v4i32 (memopv2i64 addr:$src2)))),
1727 (VSHUFPSrmi VR128:$src1, addr:$src2,
1728 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1729 // Special unary SHUFPDrri cases.
1730 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1731 (VSHUFPDrri VR128:$src1, VR128:$src1,
1732 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1733 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1734 (VSHUFPDrri VR128:$src1, VR128:$src1,
1735 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1736 // Special binary v2i64 shuffle cases using SHUFPDrri.
1737 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1738 (VSHUFPDrri VR128:$src1, VR128:$src2,
1739 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1741 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1742 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1743 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1744 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1745 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1746 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1747 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1750 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1751 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1752 def : Pat<(v8i32 (X86Shufps VR256:$src1,
1753 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
1754 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1756 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1757 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1758 def : Pat<(v8f32 (X86Shufps VR256:$src1,
1759 (memopv8f32 addr:$src2), (i8 imm:$imm))),
1760 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1762 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1763 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1764 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
1765 (memopv4i64 addr:$src2), (i8 imm:$imm))),
1766 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1768 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1769 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1770 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
1771 (memopv4f64 addr:$src2), (i8 imm:$imm))),
1772 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1775 //===----------------------------------------------------------------------===//
1776 // SSE 1 & 2 - Unpack Instructions
1777 //===----------------------------------------------------------------------===//
1779 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1780 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1781 PatFrag mem_frag, RegisterClass RC,
1782 X86MemOperand x86memop, string asm,
1784 def rr : PI<opc, MRMSrcReg,
1785 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1787 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1788 def rm : PI<opc, MRMSrcMem,
1789 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1791 (vt (OpNode RC:$src1,
1792 (mem_frag addr:$src2))))], d>;
1795 let AddedComplexity = 10 in {
1796 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1797 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1798 SSEPackedSingle>, TB, VEX_4V;
1799 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1800 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1801 SSEPackedDouble>, TB, OpSize, VEX_4V;
1802 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1803 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1804 SSEPackedSingle>, TB, VEX_4V;
1805 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1806 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1807 SSEPackedDouble>, TB, OpSize, VEX_4V;
1809 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1810 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1811 SSEPackedSingle>, TB, VEX_4V;
1812 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1813 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1814 SSEPackedDouble>, TB, OpSize, VEX_4V;
1815 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1816 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1817 SSEPackedSingle>, TB, VEX_4V;
1818 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1819 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1820 SSEPackedDouble>, TB, OpSize, VEX_4V;
1822 let Constraints = "$src1 = $dst" in {
1823 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1824 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1825 SSEPackedSingle>, TB;
1826 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1827 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1828 SSEPackedDouble>, TB, OpSize;
1829 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1830 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1831 SSEPackedSingle>, TB;
1832 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1833 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1834 SSEPackedDouble>, TB, OpSize;
1835 } // Constraints = "$src1 = $dst"
1836 } // AddedComplexity
1838 let Predicates = [HasSSE1] in {
1839 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
1840 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
1841 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
1842 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
1843 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
1844 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
1845 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
1846 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
1849 let Predicates = [HasSSE2] in {
1850 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
1851 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
1852 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
1853 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
1854 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
1855 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
1856 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
1857 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
1859 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
1860 // problem is during lowering, where it's not possible to recognize the load
1861 // fold cause it has two uses through a bitcast. One use disappears at isel
1862 // time and the fold opportunity reappears.
1863 def : Pat<(v2f64 (X86Movddup VR128:$src)),
1864 (UNPCKLPDrr VR128:$src, VR128:$src)>;
1866 let AddedComplexity = 10 in
1867 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
1868 (UNPCKLPDrr VR128:$src, VR128:$src)>;
1871 let Predicates = [HasAVX] in {
1872 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
1873 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
1874 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
1875 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
1876 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
1877 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
1878 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
1879 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
1881 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
1882 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
1883 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
1884 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
1885 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
1886 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
1887 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
1888 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
1889 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
1890 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
1891 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
1892 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
1893 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
1894 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
1895 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
1896 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
1898 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
1899 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
1900 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
1901 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
1902 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
1903 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
1904 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
1905 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
1907 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
1908 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
1909 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
1910 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
1911 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
1912 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
1913 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
1914 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
1915 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
1916 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
1917 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
1918 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
1919 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
1920 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
1921 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
1922 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
1924 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
1925 // problem is during lowering, where it's not possible to recognize the load
1926 // fold cause it has two uses through a bitcast. One use disappears at isel
1927 // time and the fold opportunity reappears.
1928 def : Pat<(v2f64 (X86Movddup VR128:$src)),
1929 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
1930 let AddedComplexity = 10 in
1931 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
1932 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
1935 //===----------------------------------------------------------------------===//
1936 // SSE 1 & 2 - Extract Floating-Point Sign mask
1937 //===----------------------------------------------------------------------===//
1939 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
1940 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
1942 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
1943 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1944 [(set GR32:$dst, (Int RC:$src))], d>;
1945 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
1946 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
1949 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
1950 SSEPackedSingle>, TB;
1951 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
1952 SSEPackedDouble>, TB, OpSize;
1954 def : Pat<(i32 (X86fgetsign FR32:$src)),
1955 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1956 sub_ss))>, Requires<[HasSSE1]>;
1957 def : Pat<(i64 (X86fgetsign FR32:$src)),
1958 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1959 sub_ss))>, Requires<[HasSSE1]>;
1960 def : Pat<(i32 (X86fgetsign FR64:$src)),
1961 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1962 sub_sd))>, Requires<[HasSSE2]>;
1963 def : Pat<(i64 (X86fgetsign FR64:$src)),
1964 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1965 sub_sd))>, Requires<[HasSSE2]>;
1967 let Predicates = [HasAVX] in {
1968 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
1969 "movmskps", SSEPackedSingle>, TB, VEX;
1970 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
1971 "movmskpd", SSEPackedDouble>, TB, OpSize,
1973 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
1974 "movmskps", SSEPackedSingle>, TB, VEX;
1975 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
1976 "movmskpd", SSEPackedDouble>, TB, OpSize,
1979 def : Pat<(i32 (X86fgetsign FR32:$src)),
1980 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1982 def : Pat<(i64 (X86fgetsign FR32:$src)),
1983 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
1985 def : Pat<(i32 (X86fgetsign FR64:$src)),
1986 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1988 def : Pat<(i64 (X86fgetsign FR64:$src)),
1989 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
1993 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1994 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
1995 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
1996 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
1998 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
1999 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2000 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2001 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
2005 //===----------------------------------------------------------------------===//
2006 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
2007 //===----------------------------------------------------------------------===//
2009 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
2010 // names that start with 'Fs'.
2012 // Alias instructions that map fld0 to pxor for sse.
2013 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
2014 canFoldAsLoad = 1 in {
2015 // FIXME: Set encoding to pseudo!
2016 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2017 [(set FR32:$dst, fp32imm0)]>,
2018 Requires<[HasSSE1]>, TB, OpSize;
2019 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2020 [(set FR64:$dst, fpimm0)]>,
2021 Requires<[HasSSE2]>, TB, OpSize;
2022 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2023 [(set FR32:$dst, fp32imm0)]>,
2024 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2025 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2026 [(set FR64:$dst, fpimm0)]>,
2027 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2030 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
2031 // bits are disregarded.
2032 let neverHasSideEffects = 1 in {
2033 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2034 "movaps\t{$src, $dst|$dst, $src}", []>;
2035 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2036 "movapd\t{$src, $dst|$dst, $src}", []>;
2039 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
2040 // bits are disregarded.
2041 let canFoldAsLoad = 1, isReMaterializable = 1 in {
2042 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
2043 "movaps\t{$src, $dst|$dst, $src}",
2044 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
2045 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
2046 "movapd\t{$src, $dst|$dst, $src}",
2047 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
2050 //===----------------------------------------------------------------------===//
2051 // SSE 1 & 2 - Logical Instructions
2052 //===----------------------------------------------------------------------===//
2054 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2056 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2058 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2059 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2061 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2062 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2064 let Constraints = "$src1 = $dst" in {
2065 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2066 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2068 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2069 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2073 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2074 let mayLoad = 0 in {
2075 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2076 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2077 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2080 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2081 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2083 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2085 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2087 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2088 // are all promoted to v2i64, and the patterns are covered by the int
2089 // version. This is needed in SSE only, because v2i64 isn't supported on
2090 // SSE1, but only on SSE2.
2091 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2092 !strconcat(OpcodeStr, "ps"), f128mem, [],
2093 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2094 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2096 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2097 !strconcat(OpcodeStr, "pd"), f128mem,
2098 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2099 (bc_v2i64 (v2f64 VR128:$src2))))],
2100 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2101 (memopv2i64 addr:$src2)))], 0>,
2103 let Constraints = "$src1 = $dst" in {
2104 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2105 !strconcat(OpcodeStr, "ps"), f128mem,
2106 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2107 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2108 (memopv2i64 addr:$src2)))]>, TB;
2110 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2111 !strconcat(OpcodeStr, "pd"), f128mem,
2112 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2113 (bc_v2i64 (v2f64 VR128:$src2))))],
2114 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2115 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2119 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2121 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2123 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2124 !strconcat(OpcodeStr, "ps"), f256mem,
2125 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2126 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2127 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2129 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2130 !strconcat(OpcodeStr, "pd"), f256mem,
2131 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2132 (bc_v4i64 (v4f64 VR256:$src2))))],
2133 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2134 (memopv4i64 addr:$src2)))], 0>,
2138 // AVX 256-bit packed logical ops forms
2139 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2140 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2141 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2142 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2144 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2145 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2146 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2147 let isCommutable = 0 in
2148 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2150 //===----------------------------------------------------------------------===//
2151 // SSE 1 & 2 - Arithmetic Instructions
2152 //===----------------------------------------------------------------------===//
2154 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2157 /// In addition, we also have a special variant of the scalar form here to
2158 /// represent the associated intrinsic operation. This form is unlike the
2159 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2160 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2162 /// These three forms can each be reg+reg or reg+mem.
2165 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2167 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2169 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2170 OpNode, FR32, f32mem, Is2Addr>, XS;
2171 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2172 OpNode, FR64, f64mem, Is2Addr>, XD;
2175 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2177 let mayLoad = 0 in {
2178 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2179 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2180 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2181 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2185 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2187 let mayLoad = 0 in {
2188 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2189 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2190 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2191 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2195 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2197 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2198 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2199 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2200 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2203 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2205 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2206 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2207 SSEPackedSingle, Is2Addr>, TB;
2209 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2210 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2211 SSEPackedDouble, Is2Addr>, TB, OpSize;
2214 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2215 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2216 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2217 SSEPackedSingle, 0>, TB;
2219 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2220 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2221 SSEPackedDouble, 0>, TB, OpSize;
2224 // Binary Arithmetic instructions
2225 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2226 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
2227 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2228 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2229 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2230 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
2231 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2232 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2234 let isCommutable = 0 in {
2235 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2236 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
2237 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2238 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2239 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2240 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
2241 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2242 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2243 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2244 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
2245 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2246 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2247 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2248 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2249 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2250 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
2251 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2252 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2253 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2254 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2257 let Constraints = "$src1 = $dst" in {
2258 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2259 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2260 basic_sse12_fp_binop_s_int<0x58, "add">;
2261 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2262 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2263 basic_sse12_fp_binop_s_int<0x59, "mul">;
2265 let isCommutable = 0 in {
2266 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2267 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2268 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2269 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2270 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2271 basic_sse12_fp_binop_s_int<0x5E, "div">;
2272 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2273 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2274 basic_sse12_fp_binop_s_int<0x5F, "max">,
2275 basic_sse12_fp_binop_p_int<0x5F, "max">;
2276 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2277 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2278 basic_sse12_fp_binop_s_int<0x5D, "min">,
2279 basic_sse12_fp_binop_p_int<0x5D, "min">;
2284 /// In addition, we also have a special variant of the scalar form here to
2285 /// represent the associated intrinsic operation. This form is unlike the
2286 /// plain scalar form, in that it takes an entire vector (instead of a
2287 /// scalar) and leaves the top elements undefined.
2289 /// And, we have a special variant form for a full-vector intrinsic form.
2291 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2292 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2293 SDNode OpNode, Intrinsic F32Int> {
2294 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2295 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2296 [(set FR32:$dst, (OpNode FR32:$src))]>;
2297 // For scalar unary operations, fold a load into the operation
2298 // only in OptForSize mode. It eliminates an instruction, but it also
2299 // eliminates a whole-register clobber (the load), so it introduces a
2300 // partial register update condition.
2301 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2302 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2303 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2304 Requires<[HasSSE1, OptForSize]>;
2305 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2306 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2307 [(set VR128:$dst, (F32Int VR128:$src))]>;
2308 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2309 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2310 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2313 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2314 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2315 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2316 !strconcat(OpcodeStr,
2317 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2318 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2319 !strconcat(OpcodeStr,
2320 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2321 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2322 (ins ssmem:$src1, VR128:$src2),
2323 !strconcat(OpcodeStr,
2324 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2327 /// sse1_fp_unop_p - SSE1 unops in packed form.
2328 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2329 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2330 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2331 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2332 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2333 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2334 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2337 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2338 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2339 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2340 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2341 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2342 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2343 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2344 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2347 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2348 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2349 Intrinsic V4F32Int> {
2350 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2351 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2352 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2353 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2354 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2355 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2358 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2359 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2360 Intrinsic V4F32Int> {
2361 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2362 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2363 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2364 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2365 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2366 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2369 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2370 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2371 SDNode OpNode, Intrinsic F64Int> {
2372 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2373 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2374 [(set FR64:$dst, (OpNode FR64:$src))]>;
2375 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2376 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2377 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2378 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2379 Requires<[HasSSE2, OptForSize]>;
2380 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2381 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2382 [(set VR128:$dst, (F64Int VR128:$src))]>;
2383 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2384 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2385 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2388 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2389 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2390 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2391 !strconcat(OpcodeStr,
2392 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2393 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2394 !strconcat(OpcodeStr,
2395 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2396 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2397 (ins VR128:$src1, sdmem:$src2),
2398 !strconcat(OpcodeStr,
2399 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2402 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2403 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2405 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2406 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2407 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2408 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2409 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2410 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2413 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2414 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2415 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2416 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2417 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2418 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2419 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2420 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2423 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2424 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2425 Intrinsic V2F64Int> {
2426 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2427 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2428 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2429 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2430 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2431 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2434 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2435 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2436 Intrinsic V2F64Int> {
2437 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2438 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2439 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2440 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2441 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2442 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2445 let Predicates = [HasAVX] in {
2447 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2448 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
2450 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2451 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2452 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2453 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2454 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2455 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2456 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2457 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2460 // Reciprocal approximations. Note that these typically require refinement
2461 // in order to obtain suitable precision.
2462 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
2463 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2464 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2465 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2466 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2468 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
2469 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2470 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2471 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2472 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2475 def : Pat<(f32 (fsqrt FR32:$src)),
2476 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2477 def : Pat<(f32 (fsqrt (load addr:$src))),
2478 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2479 Requires<[HasAVX, OptForSize]>;
2480 def : Pat<(f64 (fsqrt FR64:$src)),
2481 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2482 def : Pat<(f64 (fsqrt (load addr:$src))),
2483 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2484 Requires<[HasAVX, OptForSize]>;
2486 def : Pat<(f32 (X86frsqrt FR32:$src)),
2487 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2488 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2489 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2490 Requires<[HasAVX, OptForSize]>;
2492 def : Pat<(f32 (X86frcp FR32:$src)),
2493 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2494 def : Pat<(f32 (X86frcp (load addr:$src))),
2495 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2496 Requires<[HasAVX, OptForSize]>;
2498 let Predicates = [HasAVX] in {
2499 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2500 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2501 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2502 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2504 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2505 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2507 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2508 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2509 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2510 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2512 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2513 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2515 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2516 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2517 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2518 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2520 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2521 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2523 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2524 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2525 (VRCPSSr (f32 (IMPLICIT_DEF)),
2526 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2528 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2529 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2533 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2534 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2535 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2536 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2537 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2538 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2540 // Reciprocal approximations. Note that these typically require refinement
2541 // in order to obtain suitable precision.
2542 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2543 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2544 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2545 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2546 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2547 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2549 // There is no f64 version of the reciprocal approximation instructions.
2551 //===----------------------------------------------------------------------===//
2552 // SSE 1 & 2 - Non-temporal stores
2553 //===----------------------------------------------------------------------===//
2555 let AddedComplexity = 400 in { // Prefer non-temporal versions
2556 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2557 (ins f128mem:$dst, VR128:$src),
2558 "movntps\t{$src, $dst|$dst, $src}",
2559 [(alignednontemporalstore (v4f32 VR128:$src),
2561 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2562 (ins f128mem:$dst, VR128:$src),
2563 "movntpd\t{$src, $dst|$dst, $src}",
2564 [(alignednontemporalstore (v2f64 VR128:$src),
2566 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2567 (ins f128mem:$dst, VR128:$src),
2568 "movntdq\t{$src, $dst|$dst, $src}",
2569 [(alignednontemporalstore (v2f64 VR128:$src),
2572 let ExeDomain = SSEPackedInt in
2573 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2574 (ins f128mem:$dst, VR128:$src),
2575 "movntdq\t{$src, $dst|$dst, $src}",
2576 [(alignednontemporalstore (v4f32 VR128:$src),
2579 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2580 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
2582 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2583 (ins f256mem:$dst, VR256:$src),
2584 "movntps\t{$src, $dst|$dst, $src}",
2585 [(alignednontemporalstore (v8f32 VR256:$src),
2587 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2588 (ins f256mem:$dst, VR256:$src),
2589 "movntpd\t{$src, $dst|$dst, $src}",
2590 [(alignednontemporalstore (v4f64 VR256:$src),
2592 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2593 (ins f256mem:$dst, VR256:$src),
2594 "movntdq\t{$src, $dst|$dst, $src}",
2595 [(alignednontemporalstore (v4f64 VR256:$src),
2597 let ExeDomain = SSEPackedInt in
2598 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2599 (ins f256mem:$dst, VR256:$src),
2600 "movntdq\t{$src, $dst|$dst, $src}",
2601 [(alignednontemporalstore (v8f32 VR256:$src),
2605 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2606 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2607 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2608 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2609 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2610 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2612 let AddedComplexity = 400 in { // Prefer non-temporal versions
2613 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2614 "movntps\t{$src, $dst|$dst, $src}",
2615 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2616 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2617 "movntpd\t{$src, $dst|$dst, $src}",
2618 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2620 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2621 "movntdq\t{$src, $dst|$dst, $src}",
2622 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2624 let ExeDomain = SSEPackedInt in
2625 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2626 "movntdq\t{$src, $dst|$dst, $src}",
2627 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2629 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2630 (MOVNTDQmr addr:$dst, VR128:$src)>;
2632 // There is no AVX form for instructions below this point
2633 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2634 "movnti{l}\t{$src, $dst|$dst, $src}",
2635 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2636 TB, Requires<[HasSSE2]>;
2637 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2638 "movnti{q}\t{$src, $dst|$dst, $src}",
2639 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2640 TB, Requires<[HasSSE2]>;
2643 //===----------------------------------------------------------------------===//
2644 // SSE 1 & 2 - Prefetch and memory fence
2645 //===----------------------------------------------------------------------===//
2647 // Prefetch intrinsic.
2648 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2649 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2650 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2651 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2652 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2653 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2654 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2655 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2657 // Load, store, and memory fence
2658 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2659 TB, Requires<[HasSSE1]>;
2660 def : Pat<(X86SFence), (SFENCE)>;
2662 //===----------------------------------------------------------------------===//
2663 // SSE 1 & 2 - Load/Store XCSR register
2664 //===----------------------------------------------------------------------===//
2666 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2667 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2668 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2669 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2671 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2672 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2673 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2674 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2676 //===---------------------------------------------------------------------===//
2677 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2678 //===---------------------------------------------------------------------===//
2680 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2682 let neverHasSideEffects = 1 in {
2683 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2684 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2685 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2686 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2688 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2689 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2690 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2691 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2693 let canFoldAsLoad = 1, mayLoad = 1 in {
2694 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2695 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2696 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2697 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2698 let Predicates = [HasAVX] in {
2699 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2700 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2701 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2702 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2706 let mayStore = 1 in {
2707 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2708 (ins i128mem:$dst, VR128:$src),
2709 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2710 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2711 (ins i256mem:$dst, VR256:$src),
2712 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2713 let Predicates = [HasAVX] in {
2714 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2715 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2716 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2717 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2721 let neverHasSideEffects = 1 in
2722 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2723 "movdqa\t{$src, $dst|$dst, $src}", []>;
2725 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2726 "movdqu\t{$src, $dst|$dst, $src}",
2727 []>, XS, Requires<[HasSSE2]>;
2729 let canFoldAsLoad = 1, mayLoad = 1 in {
2730 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2731 "movdqa\t{$src, $dst|$dst, $src}",
2732 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2733 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2734 "movdqu\t{$src, $dst|$dst, $src}",
2735 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2736 XS, Requires<[HasSSE2]>;
2739 let mayStore = 1 in {
2740 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2741 "movdqa\t{$src, $dst|$dst, $src}",
2742 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2743 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2744 "movdqu\t{$src, $dst|$dst, $src}",
2745 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2746 XS, Requires<[HasSSE2]>;
2749 // Intrinsic forms of MOVDQU load and store
2750 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2751 "vmovdqu\t{$src, $dst|$dst, $src}",
2752 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2753 XS, VEX, Requires<[HasAVX]>;
2755 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2756 "movdqu\t{$src, $dst|$dst, $src}",
2757 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2758 XS, Requires<[HasSSE2]>;
2760 } // ExeDomain = SSEPackedInt
2762 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2763 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2764 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2766 //===---------------------------------------------------------------------===//
2767 // SSE2 - Packed Integer Arithmetic Instructions
2768 //===---------------------------------------------------------------------===//
2770 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2772 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2773 bit IsCommutable = 0, bit Is2Addr = 1> {
2774 let isCommutable = IsCommutable in
2775 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2776 (ins VR128:$src1, VR128:$src2),
2778 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2779 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2780 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2781 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2782 (ins VR128:$src1, i128mem:$src2),
2784 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2785 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2786 [(set VR128:$dst, (IntId VR128:$src1,
2787 (bitconvert (memopv2i64 addr:$src2))))]>;
2790 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2791 string OpcodeStr, Intrinsic IntId,
2792 Intrinsic IntId2, bit Is2Addr = 1> {
2793 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2794 (ins VR128:$src1, VR128:$src2),
2796 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2797 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2798 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2799 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2800 (ins VR128:$src1, i128mem:$src2),
2802 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2803 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2804 [(set VR128:$dst, (IntId VR128:$src1,
2805 (bitconvert (memopv2i64 addr:$src2))))]>;
2806 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2807 (ins VR128:$src1, i32i8imm:$src2),
2809 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2810 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2811 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2814 /// PDI_binop_rm - Simple SSE2 binary operator.
2815 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2816 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2817 let isCommutable = IsCommutable in
2818 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2819 (ins VR128:$src1, VR128:$src2),
2821 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2822 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2823 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2824 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2825 (ins VR128:$src1, i128mem:$src2),
2827 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2828 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2829 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2830 (bitconvert (memopv2i64 addr:$src2)))))]>;
2833 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2835 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2836 /// to collapse (bitconvert VT to VT) into its operand.
2838 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
2839 bit IsCommutable = 0, bit Is2Addr = 1> {
2840 let isCommutable = IsCommutable in
2841 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2842 (ins VR128:$src1, VR128:$src2),
2844 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2845 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2846 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
2847 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2848 (ins VR128:$src1, i128mem:$src2),
2850 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2851 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2852 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
2855 } // ExeDomain = SSEPackedInt
2857 // 128-bit Integer Arithmetic
2859 let Predicates = [HasAVX] in {
2860 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
2861 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
2862 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
2863 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
2864 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
2865 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
2866 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
2867 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
2868 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
2871 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
2873 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
2875 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
2877 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
2879 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
2881 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
2883 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
2885 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
2887 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
2889 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
2891 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
2893 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
2895 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
2897 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
2899 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
2901 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
2903 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
2905 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
2907 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
2911 let Constraints = "$src1 = $dst" in {
2912 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
2913 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
2914 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
2915 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
2916 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
2917 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
2918 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
2919 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
2920 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
2923 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
2924 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
2925 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
2926 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
2927 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
2928 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
2929 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
2930 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
2931 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
2932 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
2933 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
2934 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
2935 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
2936 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
2937 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
2938 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
2939 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
2940 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
2941 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
2943 } // Constraints = "$src1 = $dst"
2945 //===---------------------------------------------------------------------===//
2946 // SSE2 - Packed Integer Logical Instructions
2947 //===---------------------------------------------------------------------===//
2949 let Predicates = [HasAVX] in {
2950 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
2951 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
2953 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
2954 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
2956 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
2957 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
2960 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
2961 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
2963 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
2964 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
2966 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
2967 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
2970 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
2971 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
2973 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
2974 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
2977 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
2978 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
2979 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
2981 let ExeDomain = SSEPackedInt in {
2982 let neverHasSideEffects = 1 in {
2983 // 128-bit logical shifts.
2984 def VPSLLDQri : PDIi8<0x73, MRM7r,
2985 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2986 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2988 def VPSRLDQri : PDIi8<0x73, MRM3r,
2989 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
2990 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
2992 // PSRADQri doesn't exist in SSE[1-3].
2994 def VPANDNrr : PDI<0xDF, MRMSrcReg,
2995 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
2996 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2998 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3000 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3001 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3002 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3003 [(set VR128:$dst, (X86andnp VR128:$src1,
3004 (memopv2i64 addr:$src2)))]>, VEX_4V;
3008 let Constraints = "$src1 = $dst" in {
3009 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3010 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
3011 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3012 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
3013 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3014 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
3016 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3017 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
3018 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3019 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
3020 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3021 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
3023 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3024 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
3025 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3026 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
3028 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
3029 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
3030 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
3032 let ExeDomain = SSEPackedInt in {
3033 let neverHasSideEffects = 1 in {
3034 // 128-bit logical shifts.
3035 def PSLLDQri : PDIi8<0x73, MRM7r,
3036 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3037 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3038 def PSRLDQri : PDIi8<0x73, MRM3r,
3039 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3040 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3041 // PSRADQri doesn't exist in SSE[1-3].
3043 def PANDNrr : PDI<0xDF, MRMSrcReg,
3044 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3045 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3047 def PANDNrm : PDI<0xDF, MRMSrcMem,
3048 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3049 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3051 } // Constraints = "$src1 = $dst"
3053 let Predicates = [HasAVX] in {
3054 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3055 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3056 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3057 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3058 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3059 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
3060 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3061 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
3062 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3063 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3065 // Shift up / down and insert zero's.
3066 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3067 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3068 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3069 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3072 let Predicates = [HasSSE2] in {
3073 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3074 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3075 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3076 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3077 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3078 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
3079 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3080 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
3081 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3082 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3084 // Shift up / down and insert zero's.
3085 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3086 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3087 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3088 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3091 //===---------------------------------------------------------------------===//
3092 // SSE2 - Packed Integer Comparison Instructions
3093 //===---------------------------------------------------------------------===//
3095 let Predicates = [HasAVX] in {
3096 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
3098 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
3100 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
3102 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
3104 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
3106 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
3109 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3110 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3111 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3112 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3113 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3114 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3115 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3116 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3117 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3118 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3119 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3120 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3122 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3123 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3124 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3125 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3126 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3127 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3128 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3129 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3130 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3131 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3132 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3133 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3136 let Constraints = "$src1 = $dst" in {
3137 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
3138 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
3139 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
3140 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
3141 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
3142 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
3143 } // Constraints = "$src1 = $dst"
3145 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3146 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3147 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3148 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3149 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3150 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3151 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3152 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3153 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3154 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3155 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3156 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3158 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3159 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3160 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3161 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3162 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3163 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3164 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3165 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3166 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3167 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3168 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3169 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3171 //===---------------------------------------------------------------------===//
3172 // SSE2 - Packed Integer Pack Instructions
3173 //===---------------------------------------------------------------------===//
3175 let Predicates = [HasAVX] in {
3176 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3178 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3180 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3184 let Constraints = "$src1 = $dst" in {
3185 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
3186 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
3187 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
3188 } // Constraints = "$src1 = $dst"
3190 //===---------------------------------------------------------------------===//
3191 // SSE2 - Packed Integer Shuffle Instructions
3192 //===---------------------------------------------------------------------===//
3194 let ExeDomain = SSEPackedInt in {
3195 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3197 def ri : Ii8<0x70, MRMSrcReg,
3198 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3199 !strconcat(OpcodeStr,
3200 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3201 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3203 def mi : Ii8<0x70, MRMSrcMem,
3204 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3205 !strconcat(OpcodeStr,
3206 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3207 [(set VR128:$dst, (vt (pshuf_frag:$src2
3208 (bc_frag (memopv2i64 addr:$src1)),
3211 } // ExeDomain = SSEPackedInt
3213 let Predicates = [HasAVX] in {
3214 let AddedComplexity = 5 in
3215 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3218 // SSE2 with ImmT == Imm8 and XS prefix.
3219 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3222 // SSE2 with ImmT == Imm8 and XD prefix.
3223 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3226 let AddedComplexity = 5 in
3227 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3228 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3229 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3230 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3231 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3233 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3235 (VPSHUFDmi addr:$src1, imm:$imm)>;
3236 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3238 (VPSHUFDmi addr:$src1, imm:$imm)>;
3239 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3240 (VPSHUFDri VR128:$src1, imm:$imm)>;
3241 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3242 (VPSHUFDri VR128:$src1, imm:$imm)>;
3243 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3244 (VPSHUFHWri VR128:$src, imm:$imm)>;
3245 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3247 (VPSHUFHWmi addr:$src, imm:$imm)>;
3248 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3249 (VPSHUFLWri VR128:$src, imm:$imm)>;
3250 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3252 (VPSHUFLWmi addr:$src, imm:$imm)>;
3255 let Predicates = [HasSSE2] in {
3256 let AddedComplexity = 5 in
3257 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
3259 // SSE2 with ImmT == Imm8 and XS prefix.
3260 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
3262 // SSE2 with ImmT == Imm8 and XD prefix.
3263 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
3265 let AddedComplexity = 5 in
3266 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3267 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3268 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3269 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3270 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3272 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3274 (PSHUFDmi addr:$src1, imm:$imm)>;
3275 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3277 (PSHUFDmi addr:$src1, imm:$imm)>;
3278 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3279 (PSHUFDri VR128:$src1, imm:$imm)>;
3280 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3281 (PSHUFDri VR128:$src1, imm:$imm)>;
3282 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3283 (PSHUFHWri VR128:$src, imm:$imm)>;
3284 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3286 (PSHUFHWmi addr:$src, imm:$imm)>;
3287 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3288 (PSHUFLWri VR128:$src, imm:$imm)>;
3289 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3291 (PSHUFLWmi addr:$src, imm:$imm)>;
3294 //===---------------------------------------------------------------------===//
3295 // SSE2 - Packed Integer Unpack Instructions
3296 //===---------------------------------------------------------------------===//
3298 let ExeDomain = SSEPackedInt in {
3299 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3300 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3301 def rr : PDI<opc, MRMSrcReg,
3302 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3304 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3305 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3306 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3307 def rm : PDI<opc, MRMSrcMem,
3308 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3310 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3311 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3312 [(set VR128:$dst, (OpNode VR128:$src1,
3313 (bc_frag (memopv2i64
3317 let Predicates = [HasAVX] in {
3318 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
3319 bc_v16i8, 0>, VEX_4V;
3320 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
3321 bc_v8i16, 0>, VEX_4V;
3322 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
3323 bc_v4i32, 0>, VEX_4V;
3325 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3326 /// knew to collapse (bitconvert VT to VT) into its operand.
3327 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3328 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3329 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3330 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3331 VR128:$src2)))]>, VEX_4V;
3332 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3333 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3334 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3335 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3336 (memopv2i64 addr:$src2))))]>, VEX_4V;
3338 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
3339 bc_v16i8, 0>, VEX_4V;
3340 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
3341 bc_v8i16, 0>, VEX_4V;
3342 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
3343 bc_v4i32, 0>, VEX_4V;
3345 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3346 /// knew to collapse (bitconvert VT to VT) into its operand.
3347 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3348 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3349 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3350 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3351 VR128:$src2)))]>, VEX_4V;
3352 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3353 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3354 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3355 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3356 (memopv2i64 addr:$src2))))]>, VEX_4V;
3359 let Constraints = "$src1 = $dst" in {
3360 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
3361 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
3362 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
3364 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3365 /// knew to collapse (bitconvert VT to VT) into its operand.
3366 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3367 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3368 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3370 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
3371 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3372 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3373 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3375 (v2i64 (X86Punpcklqdq VR128:$src1,
3376 (memopv2i64 addr:$src2))))]>;
3378 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
3379 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
3380 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
3382 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3383 /// knew to collapse (bitconvert VT to VT) into its operand.
3384 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3385 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3386 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3388 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
3389 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3390 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3391 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3393 (v2i64 (X86Punpckhqdq VR128:$src1,
3394 (memopv2i64 addr:$src2))))]>;
3397 } // ExeDomain = SSEPackedInt
3399 //===---------------------------------------------------------------------===//
3400 // SSE2 - Packed Integer Extract and Insert
3401 //===---------------------------------------------------------------------===//
3403 let ExeDomain = SSEPackedInt in {
3404 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3405 def rri : Ii8<0xC4, MRMSrcReg,
3406 (outs VR128:$dst), (ins VR128:$src1,
3407 GR32:$src2, i32i8imm:$src3),
3409 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3410 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3412 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3413 def rmi : Ii8<0xC4, MRMSrcMem,
3414 (outs VR128:$dst), (ins VR128:$src1,
3415 i16mem:$src2, i32i8imm:$src3),
3417 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3418 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3420 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3425 let Predicates = [HasAVX] in
3426 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3427 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3428 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3429 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3430 imm:$src2))]>, TB, OpSize, VEX;
3431 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3432 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3433 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3434 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3438 let Predicates = [HasAVX] in {
3439 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
3440 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3441 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3442 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3443 []>, TB, OpSize, VEX_4V;
3446 let Constraints = "$src1 = $dst" in
3447 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3449 } // ExeDomain = SSEPackedInt
3451 //===---------------------------------------------------------------------===//
3452 // SSE2 - Packed Mask Creation
3453 //===---------------------------------------------------------------------===//
3455 let ExeDomain = SSEPackedInt in {
3457 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3458 "pmovmskb\t{$src, $dst|$dst, $src}",
3459 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
3460 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
3461 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3462 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3463 "pmovmskb\t{$src, $dst|$dst, $src}",
3464 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
3466 } // ExeDomain = SSEPackedInt
3468 //===---------------------------------------------------------------------===//
3469 // SSE2 - Conditional Store
3470 //===---------------------------------------------------------------------===//
3472 let ExeDomain = SSEPackedInt in {
3475 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
3476 (ins VR128:$src, VR128:$mask),
3477 "maskmovdqu\t{$mask, $src|$src, $mask}",
3478 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
3480 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
3481 (ins VR128:$src, VR128:$mask),
3482 "maskmovdqu\t{$mask, $src|$src, $mask}",
3483 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
3486 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3487 "maskmovdqu\t{$mask, $src|$src, $mask}",
3488 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
3490 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3491 "maskmovdqu\t{$mask, $src|$src, $mask}",
3492 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
3494 } // ExeDomain = SSEPackedInt
3496 //===---------------------------------------------------------------------===//
3497 // SSE2 - Move Doubleword
3498 //===---------------------------------------------------------------------===//
3500 //===---------------------------------------------------------------------===//
3501 // Move Int Doubleword to Packed Double Int
3503 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3504 "movd\t{$src, $dst|$dst, $src}",
3506 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
3507 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3508 "movd\t{$src, $dst|$dst, $src}",
3510 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
3512 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3513 "mov{d|q}\t{$src, $dst|$dst, $src}",
3515 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
3516 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3517 "mov{d|q}\t{$src, $dst|$dst, $src}",
3518 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
3520 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3521 "movd\t{$src, $dst|$dst, $src}",
3523 (v4i32 (scalar_to_vector GR32:$src)))]>;
3524 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3525 "movd\t{$src, $dst|$dst, $src}",
3527 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
3528 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3529 "mov{d|q}\t{$src, $dst|$dst, $src}",
3531 (v2i64 (scalar_to_vector GR64:$src)))]>;
3532 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3533 "mov{d|q}\t{$src, $dst|$dst, $src}",
3534 [(set FR64:$dst, (bitconvert GR64:$src))]>;
3536 //===---------------------------------------------------------------------===//
3537 // Move Int Doubleword to Single Scalar
3539 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3540 "movd\t{$src, $dst|$dst, $src}",
3541 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
3543 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3544 "movd\t{$src, $dst|$dst, $src}",
3545 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
3547 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3548 "movd\t{$src, $dst|$dst, $src}",
3549 [(set FR32:$dst, (bitconvert GR32:$src))]>;
3551 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3552 "movd\t{$src, $dst|$dst, $src}",
3553 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
3555 //===---------------------------------------------------------------------===//
3556 // Move Packed Doubleword Int to Packed Double Int
3558 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3559 "movd\t{$src, $dst|$dst, $src}",
3560 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3562 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
3563 (ins i32mem:$dst, VR128:$src),
3564 "movd\t{$src, $dst|$dst, $src}",
3565 [(store (i32 (vector_extract (v4i32 VR128:$src),
3566 (iPTR 0))), addr:$dst)]>, VEX;
3567 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3568 "movd\t{$src, $dst|$dst, $src}",
3569 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3571 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3572 "movd\t{$src, $dst|$dst, $src}",
3573 [(store (i32 (vector_extract (v4i32 VR128:$src),
3574 (iPTR 0))), addr:$dst)]>;
3576 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3577 "mov{d|q}\t{$src, $dst|$dst, $src}",
3578 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
3580 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
3581 "movq\t{$src, $dst|$dst, $src}",
3582 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
3584 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
3585 "mov{d|q}\t{$src, $dst|$dst, $src}",
3586 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3587 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3588 "movq\t{$src, $dst|$dst, $src}",
3589 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3591 //===---------------------------------------------------------------------===//
3592 // Move Scalar Single to Double Int
3594 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3595 "movd\t{$src, $dst|$dst, $src}",
3596 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3597 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3598 "movd\t{$src, $dst|$dst, $src}",
3599 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3600 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3601 "movd\t{$src, $dst|$dst, $src}",
3602 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3603 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3604 "movd\t{$src, $dst|$dst, $src}",
3605 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3607 //===---------------------------------------------------------------------===//
3608 // Patterns and instructions to describe movd/movq to XMM register zero-extends
3610 let AddedComplexity = 15 in {
3611 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3612 "movd\t{$src, $dst|$dst, $src}",
3613 [(set VR128:$dst, (v4i32 (X86vzmovl
3614 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3616 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3617 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3618 [(set VR128:$dst, (v2i64 (X86vzmovl
3619 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3622 let AddedComplexity = 15 in {
3623 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3624 "movd\t{$src, $dst|$dst, $src}",
3625 [(set VR128:$dst, (v4i32 (X86vzmovl
3626 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3627 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3628 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3629 [(set VR128:$dst, (v2i64 (X86vzmovl
3630 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3633 let AddedComplexity = 20 in {
3634 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3635 "movd\t{$src, $dst|$dst, $src}",
3637 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3638 (loadi32 addr:$src))))))]>,
3640 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3641 "movd\t{$src, $dst|$dst, $src}",
3643 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3644 (loadi32 addr:$src))))))]>;
3646 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3647 (MOVZDI2PDIrm addr:$src)>;
3648 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3649 (MOVZDI2PDIrm addr:$src)>;
3650 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3651 (MOVZDI2PDIrm addr:$src)>;
3654 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
3655 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
3656 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
3657 (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
3658 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
3659 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
3660 (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
3661 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
3663 // These are the correct encodings of the instructions so that we know how to
3664 // read correct assembly, even though we continue to emit the wrong ones for
3665 // compatibility with Darwin's buggy assembler.
3666 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3667 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
3668 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3669 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
3670 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3671 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
3672 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3673 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
3674 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3675 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3676 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3677 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3679 //===---------------------------------------------------------------------===//
3680 // SSE2 - Move Quadword
3681 //===---------------------------------------------------------------------===//
3683 //===---------------------------------------------------------------------===//
3684 // Move Quadword Int to Packed Quadword Int
3686 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3687 "vmovq\t{$src, $dst|$dst, $src}",
3689 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3690 VEX, Requires<[HasAVX]>;
3691 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3692 "movq\t{$src, $dst|$dst, $src}",
3694 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3695 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3697 //===---------------------------------------------------------------------===//
3698 // Move Packed Quadword Int to Quadword Int
3700 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3701 "movq\t{$src, $dst|$dst, $src}",
3702 [(store (i64 (vector_extract (v2i64 VR128:$src),
3703 (iPTR 0))), addr:$dst)]>, VEX;
3704 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3705 "movq\t{$src, $dst|$dst, $src}",
3706 [(store (i64 (vector_extract (v2i64 VR128:$src),
3707 (iPTR 0))), addr:$dst)]>;
3709 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3710 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3712 //===---------------------------------------------------------------------===//
3713 // Store / copy lower 64-bits of a XMM register.
3715 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3716 "movq\t{$src, $dst|$dst, $src}",
3717 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3718 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3719 "movq\t{$src, $dst|$dst, $src}",
3720 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3722 let AddedComplexity = 20 in
3723 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3724 "vmovq\t{$src, $dst|$dst, $src}",
3726 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3727 (loadi64 addr:$src))))))]>,
3728 XS, VEX, Requires<[HasAVX]>;
3730 let AddedComplexity = 20 in {
3731 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3732 "movq\t{$src, $dst|$dst, $src}",
3734 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3735 (loadi64 addr:$src))))))]>,
3736 XS, Requires<[HasSSE2]>;
3738 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3739 (MOVZQI2PQIrm addr:$src)>;
3740 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3741 (MOVZQI2PQIrm addr:$src)>;
3742 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3745 //===---------------------------------------------------------------------===//
3746 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3747 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3749 let AddedComplexity = 15 in
3750 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3751 "vmovq\t{$src, $dst|$dst, $src}",
3752 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3753 XS, VEX, Requires<[HasAVX]>;
3754 let AddedComplexity = 15 in
3755 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3756 "movq\t{$src, $dst|$dst, $src}",
3757 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3758 XS, Requires<[HasSSE2]>;
3760 let AddedComplexity = 20 in
3761 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3762 "vmovq\t{$src, $dst|$dst, $src}",
3763 [(set VR128:$dst, (v2i64 (X86vzmovl
3764 (loadv2i64 addr:$src))))]>,
3765 XS, VEX, Requires<[HasAVX]>;
3766 let AddedComplexity = 20 in {
3767 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3768 "movq\t{$src, $dst|$dst, $src}",
3769 [(set VR128:$dst, (v2i64 (X86vzmovl
3770 (loadv2i64 addr:$src))))]>,
3771 XS, Requires<[HasSSE2]>;
3773 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3774 (MOVZPQILo2PQIrm addr:$src)>;
3777 // Instructions to match in the assembler
3778 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3779 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3780 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3781 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3782 // Recognize "movd" with GR64 destination, but encode as a "movq"
3783 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3784 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3786 // Instructions for the disassembler
3787 // xr = XMM register
3790 let Predicates = [HasAVX] in
3791 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3792 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3793 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3794 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3796 //===---------------------------------------------------------------------===//
3797 // SSE2 - Misc Instructions
3798 //===---------------------------------------------------------------------===//
3801 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3802 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3803 TB, Requires<[HasSSE2]>;
3805 // Load, store, and memory fence
3806 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3807 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3808 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3809 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3810 def : Pat<(X86LFence), (LFENCE)>;
3811 def : Pat<(X86MFence), (MFENCE)>;
3814 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3815 // was introduced with SSE2, it's backward compatible.
3816 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3818 // Alias instructions that map zero vector to pxor / xorp* for sse.
3819 // We set canFoldAsLoad because this can be converted to a constant-pool
3820 // load of an all-ones value if folding it would be beneficial.
3821 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
3822 // JIT implementation, it does not expand the instructions below like
3823 // X86MCInstLower does.
3824 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3825 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3826 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3827 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3828 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3829 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
3830 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3831 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
3833 //===---------------------------------------------------------------------===//
3834 // SSE3 - Conversion Instructions
3835 //===---------------------------------------------------------------------===//
3837 // Convert Packed Double FP to Packed DW Integers
3838 let Predicates = [HasAVX] in {
3839 // The assembler can recognize rr 256-bit instructions by seeing a ymm
3840 // register, but the same isn't true when using memory operands instead.
3841 // Provide other assembly rr and rm forms to address this explicitly.
3842 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3843 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3844 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3845 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
3848 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3849 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3850 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3851 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
3854 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
3855 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
3856 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
3857 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
3860 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3861 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3862 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3863 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
3865 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
3866 (VCVTPD2DQYrr VR256:$src)>;
3867 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
3868 (VCVTPD2DQYrm addr:$src)>;
3870 // Convert Packed DW Integers to Packed Double FP
3871 let Predicates = [HasAVX] in {
3872 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3873 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3874 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3875 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3876 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
3877 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3878 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
3879 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
3882 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3883 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3884 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3885 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
3887 // AVX 256-bit register conversion intrinsics
3888 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
3889 (VCVTDQ2PDYrr VR128:$src)>;
3890 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
3891 (VCVTDQ2PDYrm addr:$src)>;
3893 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
3894 (VCVTPD2DQYrr VR256:$src)>;
3895 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
3896 (VCVTPD2DQYrm addr:$src)>;
3898 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
3899 (VCVTDQ2PDYrr VR128:$src)>;
3900 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
3901 (VCVTDQ2PDYrm addr:$src)>;
3903 //===---------------------------------------------------------------------===//
3904 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
3905 //===---------------------------------------------------------------------===//
3906 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
3907 ValueType vt, RegisterClass RC, PatFrag mem_frag,
3908 X86MemOperand x86memop> {
3909 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
3910 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3911 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
3912 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
3913 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3914 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
3917 let Predicates = [HasAVX] in {
3918 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3919 v4f32, VR128, memopv4f32, f128mem>, VEX;
3920 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3921 v4f32, VR128, memopv4f32, f128mem>, VEX;
3922 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
3923 v8f32, VR256, memopv8f32, f256mem>, VEX;
3924 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
3925 v8f32, VR256, memopv8f32, f256mem>, VEX;
3927 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
3928 memopv4f32, f128mem>;
3929 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
3930 memopv4f32, f128mem>;
3932 let Predicates = [HasSSE3] in {
3933 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3934 (MOVSHDUPrr VR128:$src)>;
3935 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3936 (MOVSHDUPrm addr:$src)>;
3937 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3938 (MOVSLDUPrr VR128:$src)>;
3939 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3940 (MOVSLDUPrm addr:$src)>;
3943 let Predicates = [HasAVX] in {
3944 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
3945 (VMOVSHDUPrr VR128:$src)>;
3946 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
3947 (VMOVSHDUPrm addr:$src)>;
3948 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
3949 (VMOVSLDUPrr VR128:$src)>;
3950 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
3951 (VMOVSLDUPrm addr:$src)>;
3952 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
3953 (VMOVSHDUPYrr VR256:$src)>;
3954 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
3955 (VMOVSHDUPYrm addr:$src)>;
3956 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
3957 (VMOVSLDUPYrr VR256:$src)>;
3958 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
3959 (VMOVSLDUPYrm addr:$src)>;
3962 //===---------------------------------------------------------------------===//
3963 // SSE3 - Replicate Double FP - MOVDDUP
3964 //===---------------------------------------------------------------------===//
3966 multiclass sse3_replicate_dfp<string OpcodeStr> {
3967 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3968 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3969 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
3970 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
3971 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3973 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
3977 // FIXME: Merge with above classe when there're patterns for the ymm version
3978 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
3979 let Predicates = [HasAVX] in {
3980 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3981 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3983 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3984 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
3989 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
3990 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
3991 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
3993 let Predicates = [HasSSE3] in {
3994 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
3996 (MOVDDUPrm addr:$src)>;
3997 let AddedComplexity = 5 in {
3998 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
3999 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4000 (MOVDDUPrm addr:$src)>;
4001 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4002 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4003 (MOVDDUPrm addr:$src)>;
4005 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4006 (MOVDDUPrm addr:$src)>;
4007 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4008 (MOVDDUPrm addr:$src)>;
4009 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4010 (MOVDDUPrm addr:$src)>;
4011 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4012 (MOVDDUPrm addr:$src)>;
4013 def : Pat<(X86Movddup (bc_v2f64
4014 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4015 (MOVDDUPrm addr:$src)>;
4018 let Predicates = [HasAVX] in {
4019 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4021 (VMOVDDUPrm addr:$src)>;
4022 let AddedComplexity = 5 in {
4023 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4024 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4025 (VMOVDDUPrm addr:$src)>;
4026 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4027 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4028 (VMOVDDUPrm addr:$src)>;
4030 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4031 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4032 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4033 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4034 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4035 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4036 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4037 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4038 def : Pat<(X86Movddup (bc_v2f64
4039 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4040 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4043 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4044 (VMOVDDUPYrm addr:$src)>;
4045 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4046 (VMOVDDUPYrm addr:$src)>;
4047 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4048 (VMOVDDUPYrm addr:$src)>;
4049 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4050 (VMOVDDUPYrm addr:$src)>;
4051 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4052 (VMOVDDUPYrr VR256:$src)>;
4053 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4054 (VMOVDDUPYrr VR256:$src)>;
4057 //===---------------------------------------------------------------------===//
4058 // SSE3 - Move Unaligned Integer
4059 //===---------------------------------------------------------------------===//
4061 let Predicates = [HasAVX] in {
4062 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4063 "vlddqu\t{$src, $dst|$dst, $src}",
4064 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4065 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4066 "vlddqu\t{$src, $dst|$dst, $src}",
4067 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4069 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4070 "lddqu\t{$src, $dst|$dst, $src}",
4071 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4073 //===---------------------------------------------------------------------===//
4074 // SSE3 - Arithmetic
4075 //===---------------------------------------------------------------------===//
4077 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4078 X86MemOperand x86memop, bit Is2Addr = 1> {
4079 def rr : I<0xD0, MRMSrcReg,
4080 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4082 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4083 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4084 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4085 def rm : I<0xD0, MRMSrcMem,
4086 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4088 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4089 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4090 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4093 let Predicates = [HasAVX],
4094 ExeDomain = SSEPackedDouble in {
4095 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4096 f128mem, 0>, TB, XD, VEX_4V;
4097 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4098 f128mem, 0>, TB, OpSize, VEX_4V;
4099 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4100 f256mem, 0>, TB, XD, VEX_4V;
4101 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4102 f256mem, 0>, TB, OpSize, VEX_4V;
4104 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
4105 ExeDomain = SSEPackedDouble in {
4106 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4108 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4109 f128mem>, TB, OpSize;
4112 //===---------------------------------------------------------------------===//
4113 // SSE3 Instructions
4114 //===---------------------------------------------------------------------===//
4117 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4118 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4119 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4121 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4122 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4123 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4125 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4127 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4128 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4129 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4131 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4132 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4133 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4135 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4136 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4137 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4139 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4141 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4142 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4143 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4146 let Predicates = [HasAVX] in {
4147 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4148 int_x86_sse3_hadd_ps, 0>, VEX_4V;
4149 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4150 int_x86_sse3_hadd_pd, 0>, VEX_4V;
4151 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4152 int_x86_sse3_hsub_ps, 0>, VEX_4V;
4153 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4154 int_x86_sse3_hsub_pd, 0>, VEX_4V;
4155 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4156 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
4157 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4158 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
4159 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4160 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
4161 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4162 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
4165 let Constraints = "$src1 = $dst" in {
4166 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
4167 int_x86_sse3_hadd_ps>;
4168 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
4169 int_x86_sse3_hadd_pd>;
4170 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
4171 int_x86_sse3_hsub_ps>;
4172 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
4173 int_x86_sse3_hsub_pd>;
4176 //===---------------------------------------------------------------------===//
4177 // SSSE3 - Packed Absolute Instructions
4178 //===---------------------------------------------------------------------===//
4181 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4182 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4183 PatFrag mem_frag128, Intrinsic IntId128> {
4184 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4186 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4187 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4190 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4192 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4195 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
4198 let Predicates = [HasAVX] in {
4199 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
4200 int_x86_ssse3_pabs_b_128>, VEX;
4201 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
4202 int_x86_ssse3_pabs_w_128>, VEX;
4203 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
4204 int_x86_ssse3_pabs_d_128>, VEX;
4207 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
4208 int_x86_ssse3_pabs_b_128>;
4209 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
4210 int_x86_ssse3_pabs_w_128>;
4211 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
4212 int_x86_ssse3_pabs_d_128>;
4214 //===---------------------------------------------------------------------===//
4215 // SSSE3 - Packed Binary Operator Instructions
4216 //===---------------------------------------------------------------------===//
4218 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4219 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4220 PatFrag mem_frag128, Intrinsic IntId128,
4222 let isCommutable = 1 in
4223 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4224 (ins VR128:$src1, VR128:$src2),
4226 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4227 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4228 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4230 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4231 (ins VR128:$src1, i128mem:$src2),
4233 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4234 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4236 (IntId128 VR128:$src1,
4237 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4240 let Predicates = [HasAVX] in {
4241 let isCommutable = 0 in {
4242 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
4243 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
4244 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
4245 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
4246 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
4247 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4248 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
4249 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
4250 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
4251 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
4252 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
4253 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4254 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
4255 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4256 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
4257 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
4258 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
4259 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
4260 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
4261 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
4262 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
4263 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
4265 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
4266 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4269 // None of these have i8 immediate fields.
4270 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4271 let isCommutable = 0 in {
4272 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
4273 int_x86_ssse3_phadd_w_128>;
4274 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
4275 int_x86_ssse3_phadd_d_128>;
4276 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
4277 int_x86_ssse3_phadd_sw_128>;
4278 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
4279 int_x86_ssse3_phsub_w_128>;
4280 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
4281 int_x86_ssse3_phsub_d_128>;
4282 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
4283 int_x86_ssse3_phsub_sw_128>;
4284 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
4285 int_x86_ssse3_pmadd_ub_sw_128>;
4286 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
4287 int_x86_ssse3_pshuf_b_128>;
4288 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
4289 int_x86_ssse3_psign_b_128>;
4290 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
4291 int_x86_ssse3_psign_w_128>;
4292 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
4293 int_x86_ssse3_psign_d_128>;
4295 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
4296 int_x86_ssse3_pmul_hr_sw_128>;
4299 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4300 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
4301 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4302 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
4304 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4305 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4306 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4307 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4308 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4309 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4311 //===---------------------------------------------------------------------===//
4312 // SSSE3 - Packed Align Instruction Patterns
4313 //===---------------------------------------------------------------------===//
4315 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4316 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4317 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4319 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4321 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4323 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4324 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4326 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4328 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4332 let Predicates = [HasAVX] in
4333 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4334 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4335 defm PALIGN : ssse3_palign<"palignr">;
4337 let Predicates = [HasSSSE3] in {
4338 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4339 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4340 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4341 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4342 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4343 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4344 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4345 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4348 let Predicates = [HasAVX] in {
4349 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4350 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4351 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4352 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4353 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4354 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4355 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4356 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4359 //===---------------------------------------------------------------------===//
4360 // SSSE3 Misc Instructions
4361 //===---------------------------------------------------------------------===//
4363 // Thread synchronization
4364 let usesCustomInserter = 1 in {
4365 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
4366 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
4367 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
4368 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
4371 let Uses = [EAX, ECX, EDX] in
4372 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
4373 Requires<[HasSSE3]>;
4374 let Uses = [ECX, EAX] in
4375 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
4376 Requires<[HasSSE3]>;
4378 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
4379 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
4381 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
4382 Requires<[In32BitMode]>;
4383 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
4384 Requires<[In64BitMode]>;
4386 // extload f32 -> f64. This matches load+fextend because we have a hack in
4387 // the isel (PreprocessForFPConvert) that can introduce loads after dag
4389 // Since these loads aren't folded into the fextend, we have to match it
4391 let Predicates = [HasSSE2] in
4392 def : Pat<(fextend (loadf32 addr:$src)),
4393 (CVTSS2SDrm addr:$src)>;
4395 // Move scalar to XMM zero-extended
4396 // movd to XMM register zero-extends
4397 let AddedComplexity = 15 in {
4398 // Zeroing a VR128 then do a MOVS{S|D} to the lower bits.
4399 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
4400 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
4401 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
4402 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
4403 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
4404 (MOVSSrr (v4f32 (V_SET0PS)),
4405 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
4406 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
4407 (MOVSSrr (v4i32 (V_SET0PI)),
4408 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
4411 // Splat v2f64 / v2i64
4412 let AddedComplexity = 10 in {
4413 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4414 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4417 let AddedComplexity = 20 in {
4418 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
4419 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
4420 (MOVLPSrm VR128:$src1, addr:$src2)>;
4421 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
4422 (MOVLPDrm VR128:$src1, addr:$src2)>;
4423 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
4424 (MOVLPSrm VR128:$src1, addr:$src2)>;
4425 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
4426 (MOVLPDrm VR128:$src1, addr:$src2)>;
4429 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
4430 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4431 (MOVLPSmr addr:$src1, VR128:$src2)>;
4432 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4433 (MOVLPDmr addr:$src1, VR128:$src2)>;
4434 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
4436 (MOVLPSmr addr:$src1, VR128:$src2)>;
4437 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4438 (MOVLPDmr addr:$src1, VR128:$src2)>;
4440 let AddedComplexity = 15 in {
4441 // Setting the lowest element in the vector.
4442 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
4443 (MOVSSrr (v4i32 VR128:$src1),
4444 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
4445 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
4446 (MOVSDrr (v2i64 VR128:$src1),
4447 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
4449 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
4450 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
4451 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4452 Requires<[HasSSE2]>;
4453 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
4454 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>,
4455 Requires<[HasSSE2]>;
4458 // Set lowest element and zero upper elements.
4459 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4460 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
4462 // Use movaps / movups for SSE integer load / store (one byte shorter).
4463 // The instructions selected below are then converted to MOVDQA/MOVDQU
4464 // during the SSE domain pass.
4465 let Predicates = [HasSSE1] in {
4466 def : Pat<(alignedloadv4i32 addr:$src),
4467 (MOVAPSrm addr:$src)>;
4468 def : Pat<(loadv4i32 addr:$src),
4469 (MOVUPSrm addr:$src)>;
4470 def : Pat<(alignedloadv2i64 addr:$src),
4471 (MOVAPSrm addr:$src)>;
4472 def : Pat<(loadv2i64 addr:$src),
4473 (MOVUPSrm addr:$src)>;
4475 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4476 (MOVAPSmr addr:$dst, VR128:$src)>;
4477 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4478 (MOVAPSmr addr:$dst, VR128:$src)>;
4479 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4480 (MOVAPSmr addr:$dst, VR128:$src)>;
4481 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4482 (MOVAPSmr addr:$dst, VR128:$src)>;
4483 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4484 (MOVUPSmr addr:$dst, VR128:$src)>;
4485 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4486 (MOVUPSmr addr:$dst, VR128:$src)>;
4487 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4488 (MOVUPSmr addr:$dst, VR128:$src)>;
4489 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4490 (MOVUPSmr addr:$dst, VR128:$src)>;
4493 // Use vmovaps/vmovups for AVX integer load/store.
4494 let Predicates = [HasAVX] in {
4495 // 128-bit load/store
4496 def : Pat<(alignedloadv4i32 addr:$src),
4497 (VMOVAPSrm addr:$src)>;
4498 def : Pat<(loadv4i32 addr:$src),
4499 (VMOVUPSrm addr:$src)>;
4500 def : Pat<(alignedloadv2i64 addr:$src),
4501 (VMOVAPSrm addr:$src)>;
4502 def : Pat<(loadv2i64 addr:$src),
4503 (VMOVUPSrm addr:$src)>;
4505 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4506 (VMOVAPSmr addr:$dst, VR128:$src)>;
4507 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4508 (VMOVAPSmr addr:$dst, VR128:$src)>;
4509 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4510 (VMOVAPSmr addr:$dst, VR128:$src)>;
4511 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4512 (VMOVAPSmr addr:$dst, VR128:$src)>;
4513 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4514 (VMOVUPSmr addr:$dst, VR128:$src)>;
4515 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4516 (VMOVUPSmr addr:$dst, VR128:$src)>;
4517 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4518 (VMOVUPSmr addr:$dst, VR128:$src)>;
4519 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4520 (VMOVUPSmr addr:$dst, VR128:$src)>;
4522 // 256-bit load/store
4523 def : Pat<(alignedloadv4i64 addr:$src),
4524 (VMOVAPSYrm addr:$src)>;
4525 def : Pat<(loadv4i64 addr:$src),
4526 (VMOVUPSYrm addr:$src)>;
4527 def : Pat<(alignedloadv8i32 addr:$src),
4528 (VMOVAPSYrm addr:$src)>;
4529 def : Pat<(loadv8i32 addr:$src),
4530 (VMOVUPSYrm addr:$src)>;
4531 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
4532 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4533 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
4534 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4535 def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
4536 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4537 def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
4538 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4539 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
4540 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4541 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
4542 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4543 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
4544 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4545 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
4546 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4549 //===----------------------------------------------------------------------===//
4550 // SSE4.1 - Packed Move with Sign/Zero Extend
4551 //===----------------------------------------------------------------------===//
4553 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4554 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4555 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4556 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4558 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4559 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4561 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4565 let Predicates = [HasAVX] in {
4566 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4568 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4570 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4572 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4574 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4576 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4580 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4581 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4582 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4583 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4584 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4585 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4587 // Common patterns involving scalar load.
4588 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4589 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4590 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4591 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4593 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4594 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4595 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4596 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4598 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4599 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4600 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4601 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4603 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4604 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4605 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4606 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4608 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4609 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4610 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4611 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4613 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4614 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4615 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4616 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4619 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4620 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4621 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4622 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4624 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4625 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4627 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4631 let Predicates = [HasAVX] in {
4632 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4634 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4636 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4638 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4642 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4643 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4644 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4645 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4647 // Common patterns involving scalar load
4648 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4649 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4650 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4651 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4653 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4654 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4655 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4656 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4659 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4660 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4662 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4664 // Expecting a i16 load any extended to i32 value.
4665 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4666 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4667 [(set VR128:$dst, (IntId (bitconvert
4668 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4672 let Predicates = [HasAVX] in {
4673 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4675 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4678 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4679 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4681 // Common patterns involving scalar load
4682 def : Pat<(int_x86_sse41_pmovsxbq
4683 (bitconvert (v4i32 (X86vzmovl
4684 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4685 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4687 def : Pat<(int_x86_sse41_pmovzxbq
4688 (bitconvert (v4i32 (X86vzmovl
4689 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4690 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4692 //===----------------------------------------------------------------------===//
4693 // SSE4.1 - Extract Instructions
4694 //===----------------------------------------------------------------------===//
4696 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4697 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4698 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4699 (ins VR128:$src1, i32i8imm:$src2),
4700 !strconcat(OpcodeStr,
4701 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4702 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4704 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4705 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4706 !strconcat(OpcodeStr,
4707 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4710 // There's an AssertZext in the way of writing the store pattern
4711 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4714 let Predicates = [HasAVX] in {
4715 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4716 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4717 (ins VR128:$src1, i32i8imm:$src2),
4718 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4721 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4724 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4725 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4726 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4727 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4728 !strconcat(OpcodeStr,
4729 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4732 // There's an AssertZext in the way of writing the store pattern
4733 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4736 let Predicates = [HasAVX] in
4737 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4739 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4742 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4743 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4744 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4745 (ins VR128:$src1, i32i8imm:$src2),
4746 !strconcat(OpcodeStr,
4747 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4749 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4750 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4751 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4752 !strconcat(OpcodeStr,
4753 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4754 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4755 addr:$dst)]>, OpSize;
4758 let Predicates = [HasAVX] in
4759 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4761 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4763 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4764 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4765 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4766 (ins VR128:$src1, i32i8imm:$src2),
4767 !strconcat(OpcodeStr,
4768 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4770 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4771 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4772 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4773 !strconcat(OpcodeStr,
4774 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4775 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4776 addr:$dst)]>, OpSize, REX_W;
4779 let Predicates = [HasAVX] in
4780 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4782 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4784 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4786 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4787 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4788 (ins VR128:$src1, i32i8imm:$src2),
4789 !strconcat(OpcodeStr,
4790 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4792 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4794 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4795 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4796 !strconcat(OpcodeStr,
4797 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4798 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4799 addr:$dst)]>, OpSize;
4802 let Predicates = [HasAVX] in {
4803 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4804 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4805 (ins VR128:$src1, i32i8imm:$src2),
4806 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4809 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4811 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4812 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4815 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4816 Requires<[HasSSE41]>;
4818 //===----------------------------------------------------------------------===//
4819 // SSE4.1 - Insert Instructions
4820 //===----------------------------------------------------------------------===//
4822 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4823 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4824 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4826 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4828 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4830 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4831 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4832 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4834 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4836 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4838 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4839 imm:$src3))]>, OpSize;
4842 let Predicates = [HasAVX] in
4843 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4844 let Constraints = "$src1 = $dst" in
4845 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4847 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4848 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4849 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4851 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4853 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4855 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4857 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4858 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4860 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4862 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4864 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4865 imm:$src3)))]>, OpSize;
4868 let Predicates = [HasAVX] in
4869 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4870 let Constraints = "$src1 = $dst" in
4871 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
4873 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
4874 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4875 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4877 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4879 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4881 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
4883 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4884 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
4886 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4888 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4890 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
4891 imm:$src3)))]>, OpSize;
4894 let Predicates = [HasAVX] in
4895 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
4896 let Constraints = "$src1 = $dst" in
4897 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
4899 // insertps has a few different modes, there's the first two here below which
4900 // are optimized inserts that won't zero arbitrary elements in the destination
4901 // vector. The next one matches the intrinsic and could zero arbitrary elements
4902 // in the target vector.
4903 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
4904 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4905 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
4907 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4909 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4911 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
4913 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4914 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
4916 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4918 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4920 (X86insrtps VR128:$src1,
4921 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
4922 imm:$src3))]>, OpSize;
4925 let Constraints = "$src1 = $dst" in
4926 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
4927 let Predicates = [HasAVX] in
4928 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
4930 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4931 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4933 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
4934 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
4935 Requires<[HasSSE41]>;
4937 //===----------------------------------------------------------------------===//
4938 // SSE4.1 - Round Instructions
4939 //===----------------------------------------------------------------------===//
4941 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
4942 X86MemOperand x86memop, RegisterClass RC,
4943 PatFrag mem_frag32, PatFrag mem_frag64,
4944 Intrinsic V4F32Int, Intrinsic V2F64Int> {
4945 // Intrinsic operation, reg.
4946 // Vector intrinsic operation, reg
4947 def PSr : SS4AIi8<opcps, MRMSrcReg,
4948 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4949 !strconcat(OpcodeStr,
4950 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4951 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
4954 // Vector intrinsic operation, mem
4955 def PSm : Ii8<opcps, MRMSrcMem,
4956 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4957 !strconcat(OpcodeStr,
4958 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4960 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
4962 Requires<[HasSSE41]>;
4964 // Vector intrinsic operation, reg
4965 def PDr : SS4AIi8<opcpd, MRMSrcReg,
4966 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4967 !strconcat(OpcodeStr,
4968 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4969 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
4972 // Vector intrinsic operation, mem
4973 def PDm : SS4AIi8<opcpd, MRMSrcMem,
4974 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
4975 !strconcat(OpcodeStr,
4976 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4978 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
4982 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
4983 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
4984 // Intrinsic operation, reg.
4985 // Vector intrinsic operation, reg
4986 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
4987 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
4988 !strconcat(OpcodeStr,
4989 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4992 // Vector intrinsic operation, mem
4993 def PSm_AVX : Ii8<opcps, MRMSrcMem,
4994 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
4995 !strconcat(OpcodeStr,
4996 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4997 []>, TA, OpSize, Requires<[HasSSE41]>;
4999 // Vector intrinsic operation, reg
5000 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
5001 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5002 !strconcat(OpcodeStr,
5003 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5006 // Vector intrinsic operation, mem
5007 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
5008 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5009 !strconcat(OpcodeStr,
5010 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5014 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5017 Intrinsic F64Int, bit Is2Addr = 1> {
5018 // Intrinsic operation, reg.
5019 def SSr : SS4AIi8<opcss, MRMSrcReg,
5020 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5022 !strconcat(OpcodeStr,
5023 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5024 !strconcat(OpcodeStr,
5025 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5026 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5029 // Intrinsic operation, mem.
5030 def SSm : SS4AIi8<opcss, MRMSrcMem,
5031 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5033 !strconcat(OpcodeStr,
5034 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5035 !strconcat(OpcodeStr,
5036 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5038 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5041 // Intrinsic operation, reg.
5042 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5043 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5045 !strconcat(OpcodeStr,
5046 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5047 !strconcat(OpcodeStr,
5048 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5049 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5052 // Intrinsic operation, mem.
5053 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5054 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5056 !strconcat(OpcodeStr,
5057 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5058 !strconcat(OpcodeStr,
5059 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5061 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5065 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
5067 // Intrinsic operation, reg.
5068 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
5069 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5070 !strconcat(OpcodeStr,
5071 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5074 // Intrinsic operation, mem.
5075 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
5076 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5077 !strconcat(OpcodeStr,
5078 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5081 // Intrinsic operation, reg.
5082 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
5083 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5084 !strconcat(OpcodeStr,
5085 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5088 // Intrinsic operation, mem.
5089 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
5090 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5091 !strconcat(OpcodeStr,
5092 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5096 // FP round - roundss, roundps, roundsd, roundpd
5097 let Predicates = [HasAVX] in {
5099 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5100 memopv4f32, memopv2f64,
5101 int_x86_sse41_round_ps,
5102 int_x86_sse41_round_pd>, VEX;
5103 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5104 memopv8f32, memopv4f64,
5105 int_x86_avx_round_ps_256,
5106 int_x86_avx_round_pd_256>, VEX;
5107 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5108 int_x86_sse41_round_ss,
5109 int_x86_sse41_round_sd, 0>, VEX_4V;
5111 // Instructions for the assembler
5112 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
5114 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
5116 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
5119 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5120 memopv4f32, memopv2f64,
5121 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5122 let Constraints = "$src1 = $dst" in
5123 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5124 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5126 //===----------------------------------------------------------------------===//
5127 // SSE4.1 - Packed Bit Test
5128 //===----------------------------------------------------------------------===//
5130 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5131 // the intel intrinsic that corresponds to this.
5132 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5133 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5134 "vptest\t{$src2, $src1|$src1, $src2}",
5135 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5137 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5138 "vptest\t{$src2, $src1|$src1, $src2}",
5139 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5142 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5143 "vptest\t{$src2, $src1|$src1, $src2}",
5144 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5146 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5147 "vptest\t{$src2, $src1|$src1, $src2}",
5148 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5152 let Defs = [EFLAGS] in {
5153 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5154 "ptest \t{$src2, $src1|$src1, $src2}",
5155 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5157 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5158 "ptest \t{$src2, $src1|$src1, $src2}",
5159 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5163 // The bit test instructions below are AVX only
5164 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5165 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5166 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5167 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5168 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5169 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5170 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5171 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5175 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5176 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5177 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5178 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5179 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5182 //===----------------------------------------------------------------------===//
5183 // SSE4.1 - Misc Instructions
5184 //===----------------------------------------------------------------------===//
5186 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5187 "popcnt{w}\t{$src, $dst|$dst, $src}",
5188 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
5189 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5190 "popcnt{w}\t{$src, $dst|$dst, $src}",
5191 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
5193 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5194 "popcnt{l}\t{$src, $dst|$dst, $src}",
5195 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
5196 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5197 "popcnt{l}\t{$src, $dst|$dst, $src}",
5198 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
5200 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5201 "popcnt{q}\t{$src, $dst|$dst, $src}",
5202 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
5203 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5204 "popcnt{q}\t{$src, $dst|$dst, $src}",
5205 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
5209 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5210 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5211 Intrinsic IntId128> {
5212 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5214 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5215 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5216 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5218 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5221 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
5224 let Predicates = [HasAVX] in
5225 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5226 int_x86_sse41_phminposuw>, VEX;
5227 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5228 int_x86_sse41_phminposuw>;
5230 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5231 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5232 Intrinsic IntId128, bit Is2Addr = 1> {
5233 let isCommutable = 1 in
5234 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5235 (ins VR128:$src1, VR128:$src2),
5237 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5238 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5239 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5240 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5241 (ins VR128:$src1, i128mem:$src2),
5243 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5246 (IntId128 VR128:$src1,
5247 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5250 let Predicates = [HasAVX] in {
5251 let isCommutable = 0 in
5252 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5254 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
5256 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5258 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5260 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5262 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5264 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5266 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5268 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5270 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5272 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5275 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5276 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
5277 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5278 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
5281 let Constraints = "$src1 = $dst" in {
5282 let isCommutable = 0 in
5283 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5284 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
5285 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5286 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5287 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5288 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5289 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5290 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5291 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5292 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5293 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5296 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5297 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
5298 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5299 (PCMPEQQrm VR128:$src1, addr:$src2)>;
5301 /// SS48I_binop_rm - Simple SSE41 binary operator.
5302 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5303 ValueType OpVT, bit Is2Addr = 1> {
5304 let isCommutable = 1 in
5305 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5306 (ins VR128:$src1, VR128:$src2),
5308 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5309 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5310 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
5312 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5313 (ins VR128:$src1, i128mem:$src2),
5315 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5316 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5317 [(set VR128:$dst, (OpNode VR128:$src1,
5318 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
5322 let Predicates = [HasAVX] in
5323 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
5324 let Constraints = "$src1 = $dst" in
5325 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
5327 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5328 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5329 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5330 X86MemOperand x86memop, bit Is2Addr = 1> {
5331 let isCommutable = 1 in
5332 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5333 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5335 !strconcat(OpcodeStr,
5336 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5337 !strconcat(OpcodeStr,
5338 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5339 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5341 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5342 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5344 !strconcat(OpcodeStr,
5345 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5346 !strconcat(OpcodeStr,
5347 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5350 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
5354 let Predicates = [HasAVX] in {
5355 let isCommutable = 0 in {
5356 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
5357 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5358 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
5359 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5360 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
5361 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5362 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
5363 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5364 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
5365 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5366 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
5367 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5369 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
5370 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5371 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
5372 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5373 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
5374 VR256, memopv32i8, i256mem, 0>, VEX_4V;
5377 let Constraints = "$src1 = $dst" in {
5378 let isCommutable = 0 in {
5379 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
5380 VR128, memopv16i8, i128mem>;
5381 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
5382 VR128, memopv16i8, i128mem>;
5383 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
5384 VR128, memopv16i8, i128mem>;
5385 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
5386 VR128, memopv16i8, i128mem>;
5388 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
5389 VR128, memopv16i8, i128mem>;
5390 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
5391 VR128, memopv16i8, i128mem>;
5394 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
5395 let Predicates = [HasAVX] in {
5396 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
5397 RegisterClass RC, X86MemOperand x86memop,
5398 PatFrag mem_frag, Intrinsic IntId> {
5399 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
5400 (ins RC:$src1, RC:$src2, RC:$src3),
5401 !strconcat(OpcodeStr,
5402 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5403 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
5404 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5406 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
5407 (ins RC:$src1, x86memop:$src2, RC:$src3),
5408 !strconcat(OpcodeStr,
5409 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5411 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5413 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5417 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5418 memopv16i8, int_x86_sse41_blendvpd>;
5419 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5420 memopv16i8, int_x86_sse41_blendvps>;
5421 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5422 memopv16i8, int_x86_sse41_pblendvb>;
5423 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5424 memopv32i8, int_x86_avx_blendv_pd_256>;
5425 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5426 memopv32i8, int_x86_avx_blendv_ps_256>;
5428 /// SS41I_ternary_int - SSE 4.1 ternary operator
5429 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
5430 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5431 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5432 (ins VR128:$src1, VR128:$src2),
5433 !strconcat(OpcodeStr,
5434 "\t{$src2, $dst|$dst, $src2}"),
5435 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
5438 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5439 (ins VR128:$src1, i128mem:$src2),
5440 !strconcat(OpcodeStr,
5441 "\t{$src2, $dst|$dst, $src2}"),
5444 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
5448 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
5449 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
5450 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
5452 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
5453 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
5455 let Predicates = [HasAVX] in
5456 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5457 "vmovntdqa\t{$src, $dst|$dst, $src}",
5458 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5460 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5461 "movntdqa\t{$src, $dst|$dst, $src}",
5462 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5465 //===----------------------------------------------------------------------===//
5466 // SSE4.2 - Compare Instructions
5467 //===----------------------------------------------------------------------===//
5469 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
5470 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
5471 Intrinsic IntId128, bit Is2Addr = 1> {
5472 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
5473 (ins VR128:$src1, VR128:$src2),
5475 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5476 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5477 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5479 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
5480 (ins VR128:$src1, i128mem:$src2),
5482 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5483 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5485 (IntId128 VR128:$src1,
5486 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5489 let Predicates = [HasAVX] in {
5490 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
5493 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5494 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
5495 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5496 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
5499 let Constraints = "$src1 = $dst" in
5500 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
5502 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5503 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
5504 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5505 (PCMPGTQrm VR128:$src1, addr:$src2)>;
5507 //===----------------------------------------------------------------------===//
5508 // SSE4.2 - String/text Processing Instructions
5509 //===----------------------------------------------------------------------===//
5511 // Packed Compare Implicit Length Strings, Return Mask
5512 multiclass pseudo_pcmpistrm<string asm> {
5513 def REG : PseudoI<(outs VR128:$dst),
5514 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5515 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
5517 def MEM : PseudoI<(outs VR128:$dst),
5518 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5519 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
5520 VR128:$src1, (load addr:$src2), imm:$src3))]>;
5523 let Defs = [EFLAGS], usesCustomInserter = 1 in {
5524 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
5525 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
5528 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
5529 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5530 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5531 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5532 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5533 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5534 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5537 let Defs = [XMM0, EFLAGS] in {
5538 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5539 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5540 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5541 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5542 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5543 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5546 // Packed Compare Explicit Length Strings, Return Mask
5547 multiclass pseudo_pcmpestrm<string asm> {
5548 def REG : PseudoI<(outs VR128:$dst),
5549 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5550 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5551 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5552 def MEM : PseudoI<(outs VR128:$dst),
5553 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5554 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5555 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5558 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5559 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5560 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5563 let Predicates = [HasAVX],
5564 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5565 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5566 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5567 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5568 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5569 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5570 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5573 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5574 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5575 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5576 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5577 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5578 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5579 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5582 // Packed Compare Implicit Length Strings, Return Index
5583 let Defs = [ECX, EFLAGS] in {
5584 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5585 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5586 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5587 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5588 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5589 (implicit EFLAGS)]>, OpSize;
5590 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5591 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5592 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5593 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5594 (implicit EFLAGS)]>, OpSize;
5598 let Predicates = [HasAVX] in {
5599 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5601 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5603 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5605 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5607 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5609 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5613 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5614 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5615 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5616 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5617 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5618 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5620 // Packed Compare Explicit Length Strings, Return Index
5621 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5622 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5623 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5624 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5625 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5626 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5627 (implicit EFLAGS)]>, OpSize;
5628 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5629 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5630 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5632 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5633 (implicit EFLAGS)]>, OpSize;
5637 let Predicates = [HasAVX] in {
5638 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5640 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5642 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5644 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5646 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5648 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5652 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5653 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5654 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5655 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5656 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5657 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5659 //===----------------------------------------------------------------------===//
5660 // SSE4.2 - CRC Instructions
5661 //===----------------------------------------------------------------------===//
5663 // No CRC instructions have AVX equivalents
5665 // crc intrinsic instruction
5666 // This set of instructions are only rm, the only difference is the size
5668 let Constraints = "$src1 = $dst" in {
5669 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5670 (ins GR32:$src1, i8mem:$src2),
5671 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5673 (int_x86_sse42_crc32_32_8 GR32:$src1,
5674 (load addr:$src2)))]>;
5675 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5676 (ins GR32:$src1, GR8:$src2),
5677 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5679 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5680 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5681 (ins GR32:$src1, i16mem:$src2),
5682 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5684 (int_x86_sse42_crc32_32_16 GR32:$src1,
5685 (load addr:$src2)))]>,
5687 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5688 (ins GR32:$src1, GR16:$src2),
5689 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5691 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5693 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5694 (ins GR32:$src1, i32mem:$src2),
5695 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5697 (int_x86_sse42_crc32_32_32 GR32:$src1,
5698 (load addr:$src2)))]>;
5699 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5700 (ins GR32:$src1, GR32:$src2),
5701 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5703 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5704 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5705 (ins GR64:$src1, i8mem:$src2),
5706 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5708 (int_x86_sse42_crc32_64_8 GR64:$src1,
5709 (load addr:$src2)))]>,
5711 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5712 (ins GR64:$src1, GR8:$src2),
5713 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5715 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5717 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5718 (ins GR64:$src1, i64mem:$src2),
5719 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5721 (int_x86_sse42_crc32_64_64 GR64:$src1,
5722 (load addr:$src2)))]>,
5724 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5725 (ins GR64:$src1, GR64:$src2),
5726 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5728 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5732 //===----------------------------------------------------------------------===//
5733 // AES-NI Instructions
5734 //===----------------------------------------------------------------------===//
5736 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5737 Intrinsic IntId128, bit Is2Addr = 1> {
5738 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5739 (ins VR128:$src1, VR128:$src2),
5741 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5742 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5743 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5745 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5746 (ins VR128:$src1, i128mem:$src2),
5748 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5749 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5751 (IntId128 VR128:$src1,
5752 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5755 // Perform One Round of an AES Encryption/Decryption Flow
5756 let Predicates = [HasAVX, HasAES] in {
5757 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5758 int_x86_aesni_aesenc, 0>, VEX_4V;
5759 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5760 int_x86_aesni_aesenclast, 0>, VEX_4V;
5761 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5762 int_x86_aesni_aesdec, 0>, VEX_4V;
5763 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5764 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5767 let Constraints = "$src1 = $dst" in {
5768 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5769 int_x86_aesni_aesenc>;
5770 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5771 int_x86_aesni_aesenclast>;
5772 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5773 int_x86_aesni_aesdec>;
5774 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5775 int_x86_aesni_aesdeclast>;
5778 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5779 (AESENCrr VR128:$src1, VR128:$src2)>;
5780 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5781 (AESENCrm VR128:$src1, addr:$src2)>;
5782 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5783 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5784 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5785 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5786 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5787 (AESDECrr VR128:$src1, VR128:$src2)>;
5788 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5789 (AESDECrm VR128:$src1, addr:$src2)>;
5790 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5791 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5792 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5793 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5795 // Perform the AES InvMixColumn Transformation
5796 let Predicates = [HasAVX, HasAES] in {
5797 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5799 "vaesimc\t{$src1, $dst|$dst, $src1}",
5801 (int_x86_aesni_aesimc VR128:$src1))]>,
5803 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5804 (ins i128mem:$src1),
5805 "vaesimc\t{$src1, $dst|$dst, $src1}",
5807 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5810 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5812 "aesimc\t{$src1, $dst|$dst, $src1}",
5814 (int_x86_aesni_aesimc VR128:$src1))]>,
5816 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5817 (ins i128mem:$src1),
5818 "aesimc\t{$src1, $dst|$dst, $src1}",
5820 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5823 // AES Round Key Generation Assist
5824 let Predicates = [HasAVX, HasAES] in {
5825 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5826 (ins VR128:$src1, i8imm:$src2),
5827 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5829 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5831 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5832 (ins i128mem:$src1, i8imm:$src2),
5833 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5835 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5839 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5840 (ins VR128:$src1, i8imm:$src2),
5841 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5843 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5845 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5846 (ins i128mem:$src1, i8imm:$src2),
5847 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5849 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5853 //===----------------------------------------------------------------------===//
5854 // CLMUL Instructions
5855 //===----------------------------------------------------------------------===//
5857 // Carry-less Multiplication instructions
5858 let Constraints = "$src1 = $dst" in {
5859 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5860 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5861 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5864 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5865 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5866 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5870 // AVX carry-less Multiplication instructions
5871 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5872 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5873 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5876 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5877 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5878 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5882 multiclass pclmul_alias<string asm, int immop> {
5883 def : InstAlias<!strconcat("pclmul", asm,
5884 "dq {$src, $dst|$dst, $src}"),
5885 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
5887 def : InstAlias<!strconcat("pclmul", asm,
5888 "dq {$src, $dst|$dst, $src}"),
5889 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
5891 def : InstAlias<!strconcat("vpclmul", asm,
5892 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5893 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
5895 def : InstAlias<!strconcat("vpclmul", asm,
5896 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
5897 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
5899 defm : pclmul_alias<"hqhq", 0x11>;
5900 defm : pclmul_alias<"hqlq", 0x01>;
5901 defm : pclmul_alias<"lqhq", 0x10>;
5902 defm : pclmul_alias<"lqlq", 0x00>;
5904 //===----------------------------------------------------------------------===//
5906 //===----------------------------------------------------------------------===//
5908 //===----------------------------------------------------------------------===//
5909 // VBROADCAST - Load from memory and broadcast to all elements of the
5910 // destination operand
5912 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
5913 X86MemOperand x86memop, Intrinsic Int> :
5914 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
5915 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5916 [(set RC:$dst, (Int addr:$src))]>, VEX;
5918 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
5919 int_x86_avx_vbroadcastss>;
5920 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
5921 int_x86_avx_vbroadcastss_256>;
5922 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
5923 int_x86_avx_vbroadcast_sd_256>;
5924 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
5925 int_x86_avx_vbroadcastf128_pd_256>;
5927 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
5928 (VBROADCASTF128 addr:$src)>;
5930 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
5931 (VBROADCASTSSY addr:$src)>;
5932 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
5933 (VBROADCASTSD addr:$src)>;
5934 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
5935 (VBROADCASTSSY addr:$src)>;
5936 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
5937 (VBROADCASTSD addr:$src)>;
5939 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
5940 (VBROADCASTSS addr:$src)>;
5941 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
5942 (VBROADCASTSS addr:$src)>;
5944 //===----------------------------------------------------------------------===//
5945 // VINSERTF128 - Insert packed floating-point values
5947 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
5948 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
5949 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5951 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
5952 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
5953 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
5956 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
5957 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5958 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
5959 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5960 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
5961 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
5963 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
5965 (VINSERTF128rr VR256:$src1, VR128:$src2,
5966 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5967 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
5969 (VINSERTF128rr VR256:$src1, VR128:$src2,
5970 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5971 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
5973 (VINSERTF128rr VR256:$src1, VR128:$src2,
5974 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5975 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
5977 (VINSERTF128rr VR256:$src1, VR128:$src2,
5978 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5979 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
5981 (VINSERTF128rr VR256:$src1, VR128:$src2,
5982 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5983 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
5985 (VINSERTF128rr VR256:$src1, VR128:$src2,
5986 (INSERT_get_vinsertf128_imm VR256:$ins))>;
5988 //===----------------------------------------------------------------------===//
5989 // VEXTRACTF128 - Extract packed floating-point values
5991 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
5992 (ins VR256:$src1, i8imm:$src2),
5993 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5995 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
5996 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
5997 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6000 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6001 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6002 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6003 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6004 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6005 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6007 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6008 (v4f32 (VEXTRACTF128rr
6009 (v8f32 VR256:$src1),
6010 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6011 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6012 (v2f64 (VEXTRACTF128rr
6013 (v4f64 VR256:$src1),
6014 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6015 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6016 (v4i32 (VEXTRACTF128rr
6017 (v8i32 VR256:$src1),
6018 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6019 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6020 (v2i64 (VEXTRACTF128rr
6021 (v4i64 VR256:$src1),
6022 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6023 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6024 (v8i16 (VEXTRACTF128rr
6025 (v16i16 VR256:$src1),
6026 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6027 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6028 (v16i8 (VEXTRACTF128rr
6029 (v32i8 VR256:$src1),
6030 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6032 //===----------------------------------------------------------------------===//
6033 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6035 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6036 Intrinsic IntLd, Intrinsic IntLd256,
6037 Intrinsic IntSt, Intrinsic IntSt256,
6038 PatFrag pf128, PatFrag pf256> {
6039 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6040 (ins VR128:$src1, f128mem:$src2),
6041 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6042 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6044 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6045 (ins VR256:$src1, f256mem:$src2),
6046 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6047 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6049 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6050 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6051 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6052 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6053 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6054 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6055 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6056 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6059 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6060 int_x86_avx_maskload_ps,
6061 int_x86_avx_maskload_ps_256,
6062 int_x86_avx_maskstore_ps,
6063 int_x86_avx_maskstore_ps_256,
6064 memopv4f32, memopv8f32>;
6065 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6066 int_x86_avx_maskload_pd,
6067 int_x86_avx_maskload_pd_256,
6068 int_x86_avx_maskstore_pd,
6069 int_x86_avx_maskstore_pd_256,
6070 memopv2f64, memopv4f64>;
6072 //===----------------------------------------------------------------------===//
6073 // VPERMIL - Permute Single and Double Floating-Point Values
6075 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6076 RegisterClass RC, X86MemOperand x86memop_f,
6077 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
6078 Intrinsic IntVar, Intrinsic IntImm> {
6079 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6080 (ins RC:$src1, RC:$src2),
6081 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6082 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6083 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6084 (ins RC:$src1, x86memop_i:$src2),
6085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6086 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
6088 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6089 (ins RC:$src1, i8imm:$src2),
6090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6091 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
6092 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6093 (ins x86memop_f:$src1, i8imm:$src2),
6094 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6095 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
6098 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6099 memopv4f32, memopv4i32,
6100 int_x86_avx_vpermilvar_ps,
6101 int_x86_avx_vpermil_ps>;
6102 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6103 memopv8f32, memopv8i32,
6104 int_x86_avx_vpermilvar_ps_256,
6105 int_x86_avx_vpermil_ps_256>;
6106 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6107 memopv2f64, memopv2i64,
6108 int_x86_avx_vpermilvar_pd,
6109 int_x86_avx_vpermil_pd>;
6110 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6111 memopv4f64, memopv4i64,
6112 int_x86_avx_vpermilvar_pd_256,
6113 int_x86_avx_vpermil_pd_256>;
6115 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6116 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6117 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6118 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6119 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6120 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6121 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6122 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6124 //===----------------------------------------------------------------------===//
6125 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6127 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6128 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6129 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6131 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6132 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6133 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6136 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
6137 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6138 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
6139 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6140 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
6141 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6143 def : Pat<(int_x86_avx_vperm2f128_ps_256
6144 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
6145 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6146 def : Pat<(int_x86_avx_vperm2f128_pd_256
6147 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
6148 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6149 def : Pat<(int_x86_avx_vperm2f128_si_256
6150 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
6151 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6153 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6154 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6155 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6156 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6157 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6158 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6159 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6160 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6161 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6162 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6163 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6164 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6166 //===----------------------------------------------------------------------===//
6167 // VZERO - Zero YMM registers
6169 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6170 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6171 // Zero All YMM registers
6172 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6173 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6175 // Zero Upper bits of YMM registers
6176 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6177 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
6180 //===----------------------------------------------------------------------===//
6181 // SSE Shuffle pattern fragments
6182 //===----------------------------------------------------------------------===//
6184 // This is part of a "work in progress" refactoring. The idea is that all
6185 // vector shuffles are going to be translated into target specific nodes and
6186 // directly matched by the patterns below (which can be changed along the way)
6187 // The AVX version of some but not all of them are described here, and more
6188 // should come in a near future.
6190 // Shuffle with MOVLHPD
6191 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
6192 (scalar_to_vector (loadf64 addr:$src2)))),
6193 (MOVHPDrm VR128:$src1, addr:$src2)>;
6195 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
6196 // is during lowering, where it's not possible to recognize the load fold cause
6197 // it has two uses through a bitcast. One use disappears at isel time and the
6198 // fold opportunity reappears.
6199 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
6200 (scalar_to_vector (loadf64 addr:$src2)))),
6201 (MOVHPDrm VR128:$src1, addr:$src2)>;
6203 // Shuffle with MOVSS
6204 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
6205 (MOVSSrr VR128:$src1, FR32:$src2)>;
6206 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
6207 (MOVSSrr (v4i32 VR128:$src1),
6208 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
6209 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
6210 (MOVSSrr (v4f32 VR128:$src1),
6211 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
6213 // Shuffle with MOVSD
6214 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
6215 (MOVSDrr VR128:$src1, FR64:$src2)>;
6216 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
6217 (MOVSDrr (v2i64 VR128:$src1),
6218 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
6219 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
6220 (MOVSDrr (v2f64 VR128:$src1),
6221 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
6222 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
6223 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6224 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
6225 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6227 // Shuffle with MOVLPS
6228 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
6229 (MOVLPSrm VR128:$src1, addr:$src2)>;
6230 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
6231 (MOVLPSrm VR128:$src1, addr:$src2)>;
6232 def : Pat<(X86Movlps VR128:$src1,
6233 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
6234 (MOVLPSrm VR128:$src1, addr:$src2)>;
6235 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
6236 // is during lowering, where it's not possible to recognize the load fold cause
6237 // it has two uses through a bitcast. One use disappears at isel time and the
6238 // fold opportunity reappears.
6239 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
6240 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_sd))>;
6242 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
6243 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_sd))>;
6245 // Shuffle with MOVLPD
6246 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6247 (MOVLPDrm VR128:$src1, addr:$src2)>;
6248 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6249 (MOVLPDrm VR128:$src1, addr:$src2)>;
6250 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
6251 (scalar_to_vector (loadf64 addr:$src2)))),
6252 (MOVLPDrm VR128:$src1, addr:$src2)>;
6254 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
6255 def : Pat<(store (f64 (vector_extract
6256 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6257 (MOVHPSmr addr:$dst, VR128:$src)>;
6258 def : Pat<(store (f64 (vector_extract
6259 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6260 (MOVHPDmr addr:$dst, VR128:$src)>;
6262 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
6263 (MOVLPSmr addr:$src1, VR128:$src2)>;
6264 def : Pat<(store (v4i32 (X86Movlps
6265 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
6266 (MOVLPSmr addr:$src1, VR128:$src2)>;
6268 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6269 (MOVLPDmr addr:$src1, VR128:$src2)>;
6270 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6271 (MOVLPDmr addr:$src1, VR128:$src2)>;