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 FP Scalar Instructions
300 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
301 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
302 // is used instead. Register-to-register movss/movsd is not modeled as an
303 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
304 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
305 //===----------------------------------------------------------------------===//
307 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
308 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
309 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
311 // Loading from memory automatically zeroing upper bits.
312 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
313 PatFrag mem_pat, string OpcodeStr> :
314 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
315 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
316 [(set RC:$dst, (mem_pat addr:$src))]>;
319 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
320 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V;
321 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
322 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V;
324 let canFoldAsLoad = 1, isReMaterializable = 1 in {
325 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX;
326 let AddedComplexity = 20 in
327 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX;
330 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
331 "movss\t{$src, $dst|$dst, $src}",
332 [(store FR32:$src, addr:$dst)]>, XS, VEX;
333 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
334 "movsd\t{$src, $dst|$dst, $src}",
335 [(store FR64:$src, addr:$dst)]>, XD, VEX;
338 let Constraints = "$src1 = $dst" in {
339 def MOVSSrr : sse12_move_rr<FR32, v4f32,
340 "movss\t{$src2, $dst|$dst, $src2}">, XS;
341 def MOVSDrr : sse12_move_rr<FR64, v2f64,
342 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
345 let canFoldAsLoad = 1, isReMaterializable = 1 in {
346 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
348 let AddedComplexity = 20 in
349 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
352 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
353 "movss\t{$src, $dst|$dst, $src}",
354 [(store FR32:$src, addr:$dst)]>;
355 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
356 "movsd\t{$src, $dst|$dst, $src}",
357 [(store FR64:$src, addr:$dst)]>;
360 let Predicates = [HasSSE1] in {
361 let AddedComplexity = 15 in {
362 // Extract the low 32-bit value from one vector and insert it into another.
363 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
364 (MOVSSrr (v4f32 VR128:$src1),
365 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
366 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
367 (MOVSSrr (v4i32 VR128:$src1),
368 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
370 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
371 // MOVSS to the lower bits.
372 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
373 (MOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
374 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
375 (MOVSSrr (v4f32 (V_SET0PS)),
376 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
377 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
378 (MOVSSrr (v4i32 (V_SET0PI)),
379 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
382 let AddedComplexity = 20 in {
383 // MOVSSrm zeros the high parts of the register; represent this
384 // with SUBREG_TO_REG.
385 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
386 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
387 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
388 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
389 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
390 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
393 // Extract and store.
394 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
397 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
399 // Shuffle with MOVSS
400 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
401 (MOVSSrr VR128:$src1, FR32:$src2)>;
402 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
403 (MOVSSrr (v4i32 VR128:$src1),
404 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
405 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
406 (MOVSSrr (v4f32 VR128:$src1),
407 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
410 let Predicates = [HasSSE2] in {
411 let AddedComplexity = 15 in {
412 // Extract the low 64-bit value from one vector and insert it into another.
413 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
414 (MOVSDrr (v2f64 VR128:$src1),
415 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
416 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
417 (MOVSDrr (v2i64 VR128:$src1),
418 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
420 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
421 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
422 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
423 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
424 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
426 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
427 // MOVSD to the lower bits.
428 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
429 (MOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
432 let AddedComplexity = 20 in {
433 // MOVSDrm zeros the high parts of the register; represent this
434 // with SUBREG_TO_REG.
435 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
436 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
437 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
438 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
439 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
440 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
441 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
442 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
443 def : Pat<(v2f64 (X86vzload addr:$src)),
444 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
447 // Extract and store.
448 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
451 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
453 // Shuffle with MOVSD
454 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
455 (MOVSDrr VR128:$src1, FR64:$src2)>;
456 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
457 (MOVSDrr (v2i64 VR128:$src1),
458 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
459 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
460 (MOVSDrr (v2f64 VR128:$src1),
461 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
462 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
463 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
464 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
465 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
467 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
468 // is during lowering, where it's not possible to recognize the fold cause
469 // it has two uses through a bitcast. One use disappears at isel time and the
470 // fold opportunity reappears.
471 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
472 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
473 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
474 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
477 let Predicates = [HasAVX] in {
478 let AddedComplexity = 15 in {
479 // Extract the low 32-bit value from one vector and insert it into another.
480 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
481 (VMOVSSrr (v4f32 VR128:$src1),
482 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
483 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
484 (VMOVSSrr (v4i32 VR128:$src1),
485 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
487 // Extract the low 64-bit value from one vector and insert it into another.
488 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
489 (VMOVSDrr (v2f64 VR128:$src1),
490 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
491 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
492 (VMOVSDrr (v2i64 VR128:$src1),
493 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
495 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
496 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
497 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
498 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
499 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
501 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
502 // MOVS{S,D} to the lower bits.
503 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
504 (VMOVSSrr (v4f32 (V_SET0PS)), FR32:$src)>;
505 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
506 (VMOVSSrr (v4f32 (V_SET0PS)),
507 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
508 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
509 (VMOVSSrr (v4i32 (V_SET0PI)),
510 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
511 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
512 (VMOVSDrr (v2f64 (V_SET0PS)), FR64:$src)>;
515 let AddedComplexity = 20 in {
516 // MOVSSrm zeros the high parts of the register; represent this
517 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
518 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
519 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
520 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
521 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
522 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
523 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
525 // MOVSDrm zeros the high parts of the register; represent this
526 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
527 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
528 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
529 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
530 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
531 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
532 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
533 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
534 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
535 def : Pat<(v2f64 (X86vzload addr:$src)),
536 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
538 // Represent the same patterns above but in the form they appear for
540 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
541 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
542 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
543 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
544 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
545 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
548 // Extract and store.
549 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
552 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
553 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
556 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
558 // Shuffle with VMOVSS
559 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
560 (VMOVSSrr VR128:$src1, FR32:$src2)>;
561 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
562 (VMOVSSrr (v4i32 VR128:$src1),
563 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
564 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
565 (VMOVSSrr (v4f32 VR128:$src1),
566 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
568 // Shuffle with VMOVSD
569 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
570 (VMOVSDrr VR128:$src1, FR64:$src2)>;
571 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
572 (VMOVSDrr (v2i64 VR128:$src1),
573 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
574 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
575 (VMOVSDrr (v2f64 VR128:$src1),
576 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
577 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
578 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
580 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
581 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
584 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
585 // is during lowering, where it's not possible to recognize the fold cause
586 // it has two uses through a bitcast. One use disappears at isel time and the
587 // fold opportunity reappears.
588 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
589 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
591 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
592 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
596 //===----------------------------------------------------------------------===//
597 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
598 //===----------------------------------------------------------------------===//
600 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
601 X86MemOperand x86memop, PatFrag ld_frag,
602 string asm, Domain d,
603 bit IsReMaterializable = 1> {
604 let neverHasSideEffects = 1 in
605 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
606 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
607 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
608 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
609 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
610 [(set RC:$dst, (ld_frag addr:$src))], d>;
613 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
614 "movaps", SSEPackedSingle>, TB, VEX;
615 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
616 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
617 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
618 "movups", SSEPackedSingle>, TB, VEX;
619 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
620 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
622 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
623 "movaps", SSEPackedSingle>, TB, VEX;
624 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
625 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
626 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
627 "movups", SSEPackedSingle>, TB, VEX;
628 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
629 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
630 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
631 "movaps", SSEPackedSingle>, TB;
632 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
633 "movapd", SSEPackedDouble>, TB, OpSize;
634 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
635 "movups", SSEPackedSingle>, TB;
636 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
637 "movupd", SSEPackedDouble, 0>, TB, OpSize;
639 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
640 "movaps\t{$src, $dst|$dst, $src}",
641 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
642 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
643 "movapd\t{$src, $dst|$dst, $src}",
644 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
645 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
646 "movups\t{$src, $dst|$dst, $src}",
647 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
648 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
649 "movupd\t{$src, $dst|$dst, $src}",
650 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
651 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
652 "movaps\t{$src, $dst|$dst, $src}",
653 [(alignedstore (v8f32 VR256:$src), addr:$dst)]>, VEX;
654 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
655 "movapd\t{$src, $dst|$dst, $src}",
656 [(alignedstore (v4f64 VR256:$src), addr:$dst)]>, VEX;
657 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
658 "movups\t{$src, $dst|$dst, $src}",
659 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
660 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
661 "movupd\t{$src, $dst|$dst, $src}",
662 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
664 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
665 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
666 (VMOVUPSYmr addr:$dst, VR256:$src)>;
668 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
669 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
670 (VMOVUPDYmr addr:$dst, VR256:$src)>;
672 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
673 "movaps\t{$src, $dst|$dst, $src}",
674 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
675 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
676 "movapd\t{$src, $dst|$dst, $src}",
677 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
678 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
679 "movups\t{$src, $dst|$dst, $src}",
680 [(store (v4f32 VR128:$src), addr:$dst)]>;
681 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
682 "movupd\t{$src, $dst|$dst, $src}",
683 [(store (v2f64 VR128:$src), addr:$dst)]>;
685 // Intrinsic forms of MOVUPS/D load and store
686 def VMOVUPSmr_Int : VPSI<0x11, MRMDestMem, (outs),
687 (ins f128mem:$dst, VR128:$src),
688 "movups\t{$src, $dst|$dst, $src}",
689 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>, VEX;
690 def VMOVUPDmr_Int : VPDI<0x11, MRMDestMem, (outs),
691 (ins f128mem:$dst, VR128:$src),
692 "movupd\t{$src, $dst|$dst, $src}",
693 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>, VEX;
695 def MOVUPSmr_Int : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
696 "movups\t{$src, $dst|$dst, $src}",
697 [(int_x86_sse_storeu_ps addr:$dst, VR128:$src)]>;
698 def MOVUPDmr_Int : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
699 "movupd\t{$src, $dst|$dst, $src}",
700 [(int_x86_sse2_storeu_pd addr:$dst, VR128:$src)]>;
702 // Move Low/High packed floating point values
703 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
704 PatFrag mov_frag, string base_opc,
706 def PSrm : PI<opc, MRMSrcMem,
707 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
708 !strconcat(base_opc, "s", asm_opr),
711 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
712 SSEPackedSingle>, TB;
714 def PDrm : PI<opc, MRMSrcMem,
715 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
716 !strconcat(base_opc, "d", asm_opr),
717 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
718 (scalar_to_vector (loadf64 addr:$src2)))))],
719 SSEPackedDouble>, TB, OpSize;
722 let AddedComplexity = 20 in {
723 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
724 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
725 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
726 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
728 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
729 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
730 "\t{$src2, $dst|$dst, $src2}">;
731 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
732 "\t{$src2, $dst|$dst, $src2}">;
735 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
736 "movlps\t{$src, $dst|$dst, $src}",
737 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
738 (iPTR 0))), addr:$dst)]>, VEX;
739 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
740 "movlpd\t{$src, $dst|$dst, $src}",
741 [(store (f64 (vector_extract (v2f64 VR128:$src),
742 (iPTR 0))), addr:$dst)]>, VEX;
743 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
744 "movlps\t{$src, $dst|$dst, $src}",
745 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
746 (iPTR 0))), addr:$dst)]>;
747 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
748 "movlpd\t{$src, $dst|$dst, $src}",
749 [(store (f64 (vector_extract (v2f64 VR128:$src),
750 (iPTR 0))), addr:$dst)]>;
752 // v2f64 extract element 1 is always custom lowered to unpack high to low
753 // and extract element 0 so the non-store version isn't too horrible.
754 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
755 "movhps\t{$src, $dst|$dst, $src}",
756 [(store (f64 (vector_extract
757 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
758 (undef)), (iPTR 0))), addr:$dst)]>,
760 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
761 "movhpd\t{$src, $dst|$dst, $src}",
762 [(store (f64 (vector_extract
763 (v2f64 (unpckh VR128:$src, (undef))),
764 (iPTR 0))), addr:$dst)]>,
766 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
767 "movhps\t{$src, $dst|$dst, $src}",
768 [(store (f64 (vector_extract
769 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
770 (undef)), (iPTR 0))), addr:$dst)]>;
771 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
772 "movhpd\t{$src, $dst|$dst, $src}",
773 [(store (f64 (vector_extract
774 (v2f64 (unpckh VR128:$src, (undef))),
775 (iPTR 0))), addr:$dst)]>;
777 let AddedComplexity = 20 in {
778 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
779 (ins VR128:$src1, VR128:$src2),
780 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
782 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
784 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
785 (ins VR128:$src1, VR128:$src2),
786 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
788 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
791 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
792 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
793 (ins VR128:$src1, VR128:$src2),
794 "movlhps\t{$src2, $dst|$dst, $src2}",
796 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
797 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
798 (ins VR128:$src1, VR128:$src2),
799 "movhlps\t{$src2, $dst|$dst, $src2}",
801 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
804 let Predicates = [HasAVX] in {
806 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
807 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
808 def : Pat<(X86Movlhps VR128:$src1,
809 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
810 (VMOVHPSrm VR128:$src1, addr:$src2)>;
811 def : Pat<(X86Movlhps VR128:$src1,
812 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
813 (VMOVHPSrm VR128:$src1, addr:$src2)>;
816 let AddedComplexity = 20 in {
817 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
818 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
819 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
820 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
822 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
823 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
824 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
826 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
827 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
828 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
829 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
830 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
831 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
834 let AddedComplexity = 20 in {
835 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
836 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
837 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
839 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
840 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
841 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
842 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
843 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
846 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
847 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
848 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
849 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
852 let Predicates = [HasSSE1] in {
854 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
855 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
857 def : Pat<(X86Movlhps VR128:$src1,
858 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
859 (MOVHPSrm VR128:$src1, addr:$src2)>;
860 def : Pat<(X86Movlhps VR128:$src1,
861 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
862 (MOVHPSrm VR128:$src1, addr:$src2)>;
865 let AddedComplexity = 20 in {
866 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
867 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
868 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
869 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
871 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
872 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
873 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
875 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
876 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
877 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
878 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
879 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
880 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
883 let AddedComplexity = 20 in {
884 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
885 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
886 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
888 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
889 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
890 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
891 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
892 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
895 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
896 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
897 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
898 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
901 //===----------------------------------------------------------------------===//
902 // SSE 1 & 2 - Conversion Instructions
903 //===----------------------------------------------------------------------===//
905 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
906 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
908 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
909 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
910 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
911 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
914 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
915 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
916 string asm, Domain d> {
917 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
918 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
919 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
920 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
923 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
924 X86MemOperand x86memop, string asm> {
925 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
926 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
927 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
928 (ins DstRC:$src1, x86memop:$src),
929 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
932 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
933 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX;
934 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
935 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
937 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
938 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
939 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
940 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
943 // The assembler can recognize rr 64-bit instructions by seeing a rxx
944 // register, but the same isn't true when only using memory operands,
945 // provide other assembly "l" and "q" forms to address this explicitly
946 // where appropriate to do so.
947 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
949 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
951 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
953 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
955 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
958 let Predicates = [HasAVX] in {
959 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
960 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
961 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
962 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
963 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
964 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
965 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
966 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
968 def : Pat<(f32 (sint_to_fp GR32:$src)),
969 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
970 def : Pat<(f32 (sint_to_fp GR64:$src)),
971 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
972 def : Pat<(f64 (sint_to_fp GR32:$src)),
973 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
974 def : Pat<(f64 (sint_to_fp GR64:$src)),
975 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
978 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
979 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
980 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
981 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
982 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
983 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
984 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
985 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
986 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
987 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
988 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
989 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
990 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
991 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
992 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
993 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
995 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
996 // and/or XMM operand(s).
998 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
999 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1001 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1002 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1003 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1004 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1005 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1006 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1009 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1010 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1011 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1012 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1014 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1015 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1016 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1017 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1018 (ins DstRC:$src1, x86memop:$src2),
1020 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1021 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1022 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1025 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1026 f128mem, load, "cvtsd2si">, XD, VEX;
1027 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1028 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1031 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1032 // Get rid of this hack or rename the intrinsics, there are several
1033 // intructions that only match with the intrinsic form, why create duplicates
1034 // to let them be recognized by the assembler?
1035 let Pattern = []<dag> in {
1036 defm VCVTSD2SI : sse12_cvt_s<0x2D, FR64, GR32, undef, f64mem, load,
1037 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX;
1038 defm VCVTSD2SI64 : sse12_cvt_s<0x2D, FR64, GR64, undef, f64mem, load,
1039 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W;
1041 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1042 f128mem, load, "cvtsd2si{l}">, XD;
1043 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1044 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1047 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1048 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1049 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1050 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1052 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1053 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1054 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1055 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1058 let Constraints = "$src1 = $dst" in {
1059 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1060 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1062 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1063 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1064 "cvtsi2ss{q}">, XS, REX_W;
1065 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1066 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1068 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1069 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1070 "cvtsi2sd">, XD, REX_W;
1075 // Aliases for intrinsics
1076 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1077 f32mem, load, "cvttss2si">, XS, VEX;
1078 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1079 int_x86_sse_cvttss2si64, f32mem, load,
1080 "cvttss2si">, XS, VEX, VEX_W;
1081 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1082 f128mem, load, "cvttsd2si">, XD, VEX;
1083 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1084 int_x86_sse2_cvttsd2si64, f128mem, load,
1085 "cvttsd2si">, XD, VEX, VEX_W;
1086 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1087 f32mem, load, "cvttss2si">, XS;
1088 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1089 int_x86_sse_cvttss2si64, f32mem, load,
1090 "cvttss2si{q}">, XS, REX_W;
1091 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1092 f128mem, load, "cvttsd2si">, XD;
1093 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1094 int_x86_sse2_cvttsd2si64, f128mem, load,
1095 "cvttsd2si{q}">, XD, REX_W;
1097 let Pattern = []<dag> in {
1098 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1099 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS, VEX;
1100 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1101 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1103 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1104 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1105 SSEPackedSingle>, TB, VEX;
1106 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1107 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1108 SSEPackedSingle>, TB, VEX;
1111 let Pattern = []<dag> in {
1112 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1113 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1114 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1115 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1116 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1117 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1118 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1121 let Predicates = [HasSSE1] in {
1122 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1123 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1124 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1125 (CVTSS2SIrm addr:$src)>;
1126 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1127 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1128 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1129 (CVTSS2SI64rm addr:$src)>;
1132 let Predicates = [HasAVX] in {
1133 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1134 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1135 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1136 (VCVTSS2SIrm addr:$src)>;
1137 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1138 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1139 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1140 (VCVTSS2SI64rm addr:$src)>;
1145 // Convert scalar double to scalar single
1146 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1147 (ins FR64:$src1, FR64:$src2),
1148 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1150 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1151 (ins FR64:$src1, f64mem:$src2),
1152 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1153 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V;
1154 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1157 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1158 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1159 [(set FR32:$dst, (fround FR64:$src))]>;
1160 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1161 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1162 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1163 Requires<[HasSSE2, OptForSize]>;
1165 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1166 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1168 let Constraints = "$src1 = $dst" in
1169 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1170 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1172 // Convert scalar single to scalar double
1173 // SSE2 instructions with XS prefix
1174 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1175 (ins FR32:$src1, FR32:$src2),
1176 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1177 []>, XS, Requires<[HasAVX]>, VEX_4V;
1178 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1179 (ins FR32:$src1, f32mem:$src2),
1180 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1181 []>, XS, VEX_4V, Requires<[HasAVX, OptForSize]>;
1183 let Predicates = [HasAVX] in {
1184 def : Pat<(f64 (fextend FR32:$src)),
1185 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1186 def : Pat<(fextend (loadf32 addr:$src)),
1187 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1188 def : Pat<(extloadf32 addr:$src),
1189 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1192 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1193 "cvtss2sd\t{$src, $dst|$dst, $src}",
1194 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1195 Requires<[HasSSE2]>;
1196 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1197 "cvtss2sd\t{$src, $dst|$dst, $src}",
1198 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1199 Requires<[HasSSE2, OptForSize]>;
1201 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1202 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1203 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1204 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1205 VR128:$src2))]>, XS, VEX_4V,
1207 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1208 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1209 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1210 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1211 (load addr:$src2)))]>, XS, VEX_4V,
1213 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1214 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1215 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1216 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1217 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1218 VR128:$src2))]>, XS,
1219 Requires<[HasSSE2]>;
1220 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1221 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1222 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1223 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1224 (load addr:$src2)))]>, XS,
1225 Requires<[HasSSE2]>;
1228 def : Pat<(extloadf32 addr:$src),
1229 (CVTSS2SDrr (MOVSSrm addr:$src))>,
1230 Requires<[HasSSE2, OptForSpeed]>;
1232 // Convert doubleword to packed single/double fp
1233 // SSE2 instructions without OpSize prefix
1234 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1235 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1236 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1237 TB, VEX, Requires<[HasAVX]>;
1238 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1239 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1240 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1241 (bitconvert (memopv2i64 addr:$src))))]>,
1242 TB, VEX, Requires<[HasAVX]>;
1243 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1244 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1245 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1246 TB, Requires<[HasSSE2]>;
1247 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1248 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1249 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1250 (bitconvert (memopv2i64 addr:$src))))]>,
1251 TB, Requires<[HasSSE2]>;
1253 // FIXME: why the non-intrinsic version is described as SSE3?
1254 // SSE2 instructions with XS prefix
1255 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1256 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1257 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1258 XS, VEX, Requires<[HasAVX]>;
1259 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1260 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1261 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1262 (bitconvert (memopv2i64 addr:$src))))]>,
1263 XS, VEX, Requires<[HasAVX]>;
1264 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1265 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1266 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1267 XS, Requires<[HasSSE2]>;
1268 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1269 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1270 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1271 (bitconvert (memopv2i64 addr:$src))))]>,
1272 XS, Requires<[HasSSE2]>;
1275 // Convert packed single/double fp to doubleword
1276 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1277 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1278 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1279 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1280 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1281 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1282 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1283 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1284 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1285 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1286 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1287 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1289 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1290 "cvtps2dq\t{$src, $dst|$dst, $src}",
1291 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1293 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1295 "cvtps2dq\t{$src, $dst|$dst, $src}",
1296 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1297 (memop addr:$src)))]>, VEX;
1298 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1299 "cvtps2dq\t{$src, $dst|$dst, $src}",
1300 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1301 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1302 "cvtps2dq\t{$src, $dst|$dst, $src}",
1303 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1304 (memop addr:$src)))]>;
1306 // SSE2 packed instructions with XD prefix
1307 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1308 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1309 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1310 XD, VEX, Requires<[HasAVX]>;
1311 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1312 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1313 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1314 (memop addr:$src)))]>,
1315 XD, VEX, Requires<[HasAVX]>;
1316 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1317 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1318 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1319 XD, Requires<[HasSSE2]>;
1320 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1321 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1322 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1323 (memop addr:$src)))]>,
1324 XD, Requires<[HasSSE2]>;
1327 // Convert with truncation packed single/double fp to doubleword
1328 // SSE2 packed instructions with XS prefix
1329 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1330 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1331 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1332 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1333 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1334 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1335 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1336 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1337 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1338 "cvttps2dq\t{$src, $dst|$dst, $src}",
1340 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1341 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1342 "cvttps2dq\t{$src, $dst|$dst, $src}",
1344 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1346 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1347 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1349 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1350 XS, VEX, Requires<[HasAVX]>;
1351 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1352 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1353 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1354 (memop addr:$src)))]>,
1355 XS, VEX, Requires<[HasAVX]>;
1357 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1358 (Int_CVTDQ2PSrr VR128:$src)>, Requires<[HasSSE2]>;
1359 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1360 (CVTTPS2DQrr VR128:$src)>, Requires<[HasSSE2]>;
1362 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1363 (Int_VCVTDQ2PSrr VR128:$src)>, Requires<[HasAVX]>;
1364 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1365 (VCVTTPS2DQrr VR128:$src)>, Requires<[HasAVX]>;
1366 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1367 (VCVTDQ2PSYrr VR256:$src)>, Requires<[HasAVX]>;
1368 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1369 (VCVTTPS2DQYrr VR256:$src)>, Requires<[HasAVX]>;
1371 def Int_VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst),
1373 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1374 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>,
1376 def Int_VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst),
1378 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1379 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1380 (memop addr:$src)))]>, VEX;
1381 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1382 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1383 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1384 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1385 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1386 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1387 (memop addr:$src)))]>;
1389 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1390 // register, but the same isn't true when using memory operands instead.
1391 // Provide other assembly rr and rm forms to address this explicitly.
1392 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1393 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1394 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1395 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1398 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1399 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1400 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1401 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1404 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1405 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1406 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1407 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1409 // Convert packed single to packed double
1410 let Predicates = [HasAVX] in {
1411 // SSE2 instructions without OpSize prefix
1412 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1413 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1414 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1415 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1416 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1417 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1418 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1419 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1421 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1422 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1423 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1424 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1426 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1427 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1428 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1429 TB, VEX, Requires<[HasAVX]>;
1430 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1431 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1432 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1433 (load addr:$src)))]>,
1434 TB, VEX, Requires<[HasAVX]>;
1435 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1436 "cvtps2pd\t{$src, $dst|$dst, $src}",
1437 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1438 TB, Requires<[HasSSE2]>;
1439 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1440 "cvtps2pd\t{$src, $dst|$dst, $src}",
1441 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1442 (load addr:$src)))]>,
1443 TB, Requires<[HasSSE2]>;
1445 // Convert packed double to packed single
1446 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1447 // register, but the same isn't true when using memory operands instead.
1448 // Provide other assembly rr and rm forms to address this explicitly.
1449 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1450 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1451 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1452 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1455 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1456 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1457 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1458 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1461 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1462 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1463 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1464 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1465 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1466 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1467 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1468 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1471 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1472 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1473 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1474 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1476 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1477 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1478 (memop addr:$src)))]>;
1479 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1480 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1481 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1482 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1483 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1484 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1485 (memop addr:$src)))]>;
1487 // AVX 256-bit register conversion intrinsics
1488 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1489 // whenever possible to avoid declaring two versions of each one.
1490 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1491 (VCVTDQ2PSYrr VR256:$src)>;
1492 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1493 (VCVTDQ2PSYrm addr:$src)>;
1495 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1496 (VCVTPD2PSYrr VR256:$src)>;
1497 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1498 (VCVTPD2PSYrm addr:$src)>;
1500 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1501 (VCVTPS2DQYrr VR256:$src)>;
1502 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1503 (VCVTPS2DQYrm addr:$src)>;
1505 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1506 (VCVTPS2PDYrr VR128:$src)>;
1507 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1508 (VCVTPS2PDYrm addr:$src)>;
1510 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1511 (VCVTTPD2DQYrr VR256:$src)>;
1512 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1513 (VCVTTPD2DQYrm addr:$src)>;
1515 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1516 (VCVTTPS2DQYrr VR256:$src)>;
1517 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1518 (VCVTTPS2DQYrm addr:$src)>;
1520 // Match fround and fextend for 128/256-bit conversions
1521 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1522 (VCVTPD2PSYrr VR256:$src)>;
1523 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1524 (VCVTPD2PSYrm addr:$src)>;
1526 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1527 (VCVTPS2PDYrr VR128:$src)>;
1528 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1529 (VCVTPS2PDYrm addr:$src)>;
1531 //===----------------------------------------------------------------------===//
1532 // SSE 1 & 2 - Compare Instructions
1533 //===----------------------------------------------------------------------===//
1535 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1536 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1537 string asm, string asm_alt> {
1538 let isAsmParserOnly = 1 in {
1539 def rr : SIi8<0xC2, MRMSrcReg,
1540 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc),
1543 def rm : SIi8<0xC2, MRMSrcMem,
1544 (outs RC:$dst), (ins RC:$src1, x86memop:$src, SSECC:$cc),
1548 // Accept explicit immediate argument form instead of comparison code.
1549 def rr_alt : SIi8<0xC2, MRMSrcReg,
1550 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1553 def rm_alt : SIi8<0xC2, MRMSrcMem,
1554 (outs RC:$dst), (ins RC:$src1, x86memop:$src, i8imm:$src2),
1558 let neverHasSideEffects = 1 in {
1559 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem,
1560 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}",
1561 "cmpss\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1563 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem,
1564 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}",
1565 "cmpsd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}">,
1569 let Constraints = "$src1 = $dst" in {
1570 def CMPSSrr : SIi8<0xC2, MRMSrcReg,
1571 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, SSECC:$cc),
1572 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1573 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), FR32:$src2, imm:$cc))]>, XS;
1574 def CMPSSrm : SIi8<0xC2, MRMSrcMem,
1575 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src2, SSECC:$cc),
1576 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
1577 [(set FR32:$dst, (X86cmpss (f32 FR32:$src1), (loadf32 addr:$src2), imm:$cc))]>, XS;
1578 def CMPSDrr : SIi8<0xC2, MRMSrcReg,
1579 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, SSECC:$cc),
1580 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1581 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), FR64:$src2, imm:$cc))]>, XD;
1582 def CMPSDrm : SIi8<0xC2, MRMSrcMem,
1583 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src2, SSECC:$cc),
1584 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
1585 [(set FR64:$dst, (X86cmpsd (f64 FR64:$src1), (loadf64 addr:$src2), imm:$cc))]>, XD;
1587 let Constraints = "$src1 = $dst", neverHasSideEffects = 1 in {
1588 def CMPSSrr_alt : SIi8<0xC2, MRMSrcReg,
1589 (outs FR32:$dst), (ins FR32:$src1, FR32:$src, i8imm:$src2),
1590 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1591 def CMPSSrm_alt : SIi8<0xC2, MRMSrcMem,
1592 (outs FR32:$dst), (ins FR32:$src1, f32mem:$src, i8imm:$src2),
1593 "cmpss\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XS;
1594 def CMPSDrr_alt : SIi8<0xC2, MRMSrcReg,
1595 (outs FR64:$dst), (ins FR64:$src1, FR64:$src, i8imm:$src2),
1596 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1597 def CMPSDrm_alt : SIi8<0xC2, MRMSrcMem,
1598 (outs FR64:$dst), (ins FR64:$src1, f64mem:$src, i8imm:$src2),
1599 "cmpsd\t{$src2, $src, $dst|$dst, $src, $src2}", []>, XD;
1602 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
1603 Intrinsic Int, string asm> {
1604 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
1605 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
1606 [(set VR128:$dst, (Int VR128:$src1,
1607 VR128:$src, imm:$cc))]>;
1608 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
1609 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
1610 [(set VR128:$dst, (Int VR128:$src1,
1611 (load addr:$src), imm:$cc))]>;
1614 // Aliases to match intrinsics which expect XMM operand(s).
1615 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1616 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
1618 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1619 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
1621 let Constraints = "$src1 = $dst" in {
1622 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
1623 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
1624 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
1625 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
1629 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
1630 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
1631 ValueType vt, X86MemOperand x86memop,
1632 PatFrag ld_frag, string OpcodeStr, Domain d> {
1633 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
1634 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1635 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
1636 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
1637 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
1638 [(set EFLAGS, (OpNode (vt RC:$src1),
1639 (ld_frag addr:$src2)))], d>;
1642 let Defs = [EFLAGS] in {
1643 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1644 "ucomiss", SSEPackedSingle>, TB, VEX;
1645 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1646 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1647 let Pattern = []<dag> in {
1648 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1649 "comiss", SSEPackedSingle>, TB, VEX;
1650 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1651 "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1654 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1655 load, "ucomiss", SSEPackedSingle>, TB, VEX;
1656 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1657 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
1659 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
1660 load, "comiss", SSEPackedSingle>, TB, VEX;
1661 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
1662 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
1663 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
1664 "ucomiss", SSEPackedSingle>, TB;
1665 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
1666 "ucomisd", SSEPackedDouble>, TB, OpSize;
1668 let Pattern = []<dag> in {
1669 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
1670 "comiss", SSEPackedSingle>, TB;
1671 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
1672 "comisd", SSEPackedDouble>, TB, OpSize;
1675 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
1676 load, "ucomiss", SSEPackedSingle>, TB;
1677 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
1678 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
1680 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
1681 "comiss", SSEPackedSingle>, TB;
1682 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
1683 "comisd", SSEPackedDouble>, TB, OpSize;
1684 } // Defs = [EFLAGS]
1686 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
1687 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
1688 Intrinsic Int, string asm, string asm_alt,
1690 let isAsmParserOnly = 1 in {
1691 def rri : PIi8<0xC2, MRMSrcReg,
1692 (outs RC:$dst), (ins RC:$src1, RC:$src, SSECC:$cc), asm,
1693 [(set RC:$dst, (Int RC:$src1, RC:$src, imm:$cc))], d>;
1694 def rmi : PIi8<0xC2, MRMSrcMem,
1695 (outs RC:$dst), (ins RC:$src1, f128mem:$src, SSECC:$cc), asm,
1696 [(set RC:$dst, (Int RC:$src1, (memop addr:$src), imm:$cc))], d>;
1699 // Accept explicit immediate argument form instead of comparison code.
1700 def rri_alt : PIi8<0xC2, MRMSrcReg,
1701 (outs RC:$dst), (ins RC:$src1, RC:$src, i8imm:$src2),
1703 def rmi_alt : PIi8<0xC2, MRMSrcMem,
1704 (outs RC:$dst), (ins RC:$src1, f128mem:$src, i8imm:$src2),
1708 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1709 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1710 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1711 SSEPackedSingle>, TB, VEX_4V;
1712 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1713 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1714 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1715 SSEPackedDouble>, TB, OpSize, VEX_4V;
1716 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
1717 "cmp${cc}ps\t{$src, $src1, $dst|$dst, $src1, $src}",
1718 "cmpps\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1719 SSEPackedSingle>, TB, VEX_4V;
1720 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
1721 "cmp${cc}pd\t{$src, $src1, $dst|$dst, $src1, $src}",
1722 "cmppd\t{$src2, $src, $src1, $dst|$dst, $src1, $src, $src2}",
1723 SSEPackedDouble>, TB, OpSize, VEX_4V;
1724 let Constraints = "$src1 = $dst" in {
1725 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
1726 "cmp${cc}ps\t{$src, $dst|$dst, $src}",
1727 "cmpps\t{$src2, $src, $dst|$dst, $src, $src2}",
1728 SSEPackedSingle>, TB;
1729 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
1730 "cmp${cc}pd\t{$src, $dst|$dst, $src}",
1731 "cmppd\t{$src2, $src, $dst|$dst, $src, $src2}",
1732 SSEPackedDouble>, TB, OpSize;
1735 let Predicates = [HasSSE1] in {
1736 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1737 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1738 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1739 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1742 let Predicates = [HasSSE2] in {
1743 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1744 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1745 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1746 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1749 let Predicates = [HasAVX] in {
1750 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
1751 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
1752 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
1753 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
1754 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
1755 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
1756 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
1757 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
1759 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
1760 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
1761 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
1762 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
1763 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
1764 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
1765 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
1766 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
1769 //===----------------------------------------------------------------------===//
1770 // SSE 1 & 2 - Shuffle Instructions
1771 //===----------------------------------------------------------------------===//
1773 /// sse12_shuffle - sse 1 & 2 shuffle instructions
1774 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
1775 ValueType vt, string asm, PatFrag mem_frag,
1776 Domain d, bit IsConvertibleToThreeAddress = 0> {
1777 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
1778 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
1779 [(set RC:$dst, (vt (shufp:$src3
1780 RC:$src1, (mem_frag addr:$src2))))], d>;
1781 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
1782 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
1783 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
1785 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
1788 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1789 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1790 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
1791 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
1792 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
1793 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
1794 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1795 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1796 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1797 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
1798 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
1799 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
1801 let Constraints = "$src1 = $dst" in {
1802 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
1803 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1804 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
1806 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
1807 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
1808 memopv2f64, SSEPackedDouble>, TB, OpSize;
1811 let Predicates = [HasSSE1] in {
1812 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1813 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1814 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1815 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1816 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1817 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1818 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1819 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1820 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1821 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1822 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1823 // fall back to this for SSE1)
1824 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1825 (SHUFPSrri VR128:$src2, VR128:$src1,
1826 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1827 // Special unary SHUFPSrri case.
1828 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1829 (SHUFPSrri VR128:$src1, VR128:$src1,
1830 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1833 let Predicates = [HasSSE2] in {
1834 // Special binary v4i32 shuffle cases with SHUFPS.
1835 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1836 (SHUFPSrri VR128:$src1, VR128:$src2,
1837 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1838 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1839 (bc_v4i32 (memopv2i64 addr:$src2)))),
1840 (SHUFPSrmi VR128:$src1, addr:$src2,
1841 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1842 // Special unary SHUFPDrri cases.
1843 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1844 (SHUFPDrri VR128:$src1, VR128:$src1,
1845 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1846 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1847 (SHUFPDrri VR128:$src1, VR128:$src1,
1848 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1849 // Special binary v2i64 shuffle cases using SHUFPDrri.
1850 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1851 (SHUFPDrri VR128:$src1, VR128:$src2,
1852 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1853 // Generic SHUFPD patterns
1854 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1855 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1856 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1857 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1858 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1859 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1860 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1863 let Predicates = [HasAVX] in {
1864 def : Pat<(v4f32 (X86Shufps VR128:$src1,
1865 (memopv4f32 addr:$src2), (i8 imm:$imm))),
1866 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1867 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1868 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1869 def : Pat<(v4i32 (X86Shufps VR128:$src1,
1870 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
1871 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
1872 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1873 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
1874 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
1875 // fall back to this for SSE1)
1876 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
1877 (VSHUFPSrri VR128:$src2, VR128:$src1,
1878 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1879 // Special unary SHUFPSrri case.
1880 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
1881 (VSHUFPSrri VR128:$src1, VR128:$src1,
1882 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1883 // Special binary v4i32 shuffle cases with SHUFPS.
1884 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
1885 (VSHUFPSrri VR128:$src1, VR128:$src2,
1886 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1887 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
1888 (bc_v4i32 (memopv2i64 addr:$src2)))),
1889 (VSHUFPSrmi VR128:$src1, addr:$src2,
1890 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1891 // Special unary SHUFPDrri cases.
1892 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
1893 (VSHUFPDrri VR128:$src1, VR128:$src1,
1894 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1895 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
1896 (VSHUFPDrri VR128:$src1, VR128:$src1,
1897 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1898 // Special binary v2i64 shuffle cases using SHUFPDrri.
1899 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
1900 (VSHUFPDrri VR128:$src1, VR128:$src2,
1901 (SHUFFLE_get_shuf_imm VR128:$src3))>;
1903 def : Pat<(v2f64 (X86Shufps VR128:$src1,
1904 (memopv2f64 addr:$src2), (i8 imm:$imm))),
1905 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
1906 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1907 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1908 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
1909 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
1912 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1913 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1914 def : Pat<(v8i32 (X86Shufps VR256:$src1,
1915 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
1916 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1918 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1919 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1920 def : Pat<(v8f32 (X86Shufps VR256:$src1,
1921 (memopv8f32 addr:$src2), (i8 imm:$imm))),
1922 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1924 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1925 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1926 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
1927 (memopv4i64 addr:$src2), (i8 imm:$imm))),
1928 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1930 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
1931 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
1932 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
1933 (memopv4f64 addr:$src2), (i8 imm:$imm))),
1934 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
1937 //===----------------------------------------------------------------------===//
1938 // SSE 1 & 2 - Unpack Instructions
1939 //===----------------------------------------------------------------------===//
1941 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
1942 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
1943 PatFrag mem_frag, RegisterClass RC,
1944 X86MemOperand x86memop, string asm,
1946 def rr : PI<opc, MRMSrcReg,
1947 (outs RC:$dst), (ins RC:$src1, RC:$src2),
1949 (vt (OpNode RC:$src1, RC:$src2)))], d>;
1950 def rm : PI<opc, MRMSrcMem,
1951 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
1953 (vt (OpNode RC:$src1,
1954 (mem_frag addr:$src2))))], d>;
1957 let AddedComplexity = 10 in {
1958 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1959 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1960 SSEPackedSingle>, TB, VEX_4V;
1961 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1962 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1963 SSEPackedDouble>, TB, OpSize, VEX_4V;
1964 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1965 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1966 SSEPackedSingle>, TB, VEX_4V;
1967 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1968 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1969 SSEPackedDouble>, TB, OpSize, VEX_4V;
1971 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
1972 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1973 SSEPackedSingle>, TB, VEX_4V;
1974 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
1975 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1976 SSEPackedDouble>, TB, OpSize, VEX_4V;
1977 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
1978 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1979 SSEPackedSingle>, TB, VEX_4V;
1980 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
1981 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1982 SSEPackedDouble>, TB, OpSize, VEX_4V;
1984 let Constraints = "$src1 = $dst" in {
1985 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
1986 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
1987 SSEPackedSingle>, TB;
1988 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
1989 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
1990 SSEPackedDouble>, TB, OpSize;
1991 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
1992 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
1993 SSEPackedSingle>, TB;
1994 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
1995 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
1996 SSEPackedDouble>, TB, OpSize;
1997 } // Constraints = "$src1 = $dst"
1998 } // AddedComplexity
2000 let Predicates = [HasSSE1] in {
2001 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2002 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2003 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2004 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2005 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2006 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2007 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2008 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2011 let Predicates = [HasSSE2] in {
2012 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2013 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2014 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2015 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2016 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2017 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2018 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2019 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2021 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2022 // problem is during lowering, where it's not possible to recognize the load
2023 // fold cause it has two uses through a bitcast. One use disappears at isel
2024 // time and the fold opportunity reappears.
2025 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2026 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2028 let AddedComplexity = 10 in
2029 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2030 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2033 let Predicates = [HasAVX] in {
2034 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, (memopv4f32 addr:$src2))),
2035 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2036 def : Pat<(v4f32 (X86Unpcklps VR128:$src1, VR128:$src2)),
2037 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2038 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, (memopv4f32 addr:$src2))),
2039 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2040 def : Pat<(v4f32 (X86Unpckhps VR128:$src1, VR128:$src2)),
2041 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2043 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, (memopv8f32 addr:$src2))),
2044 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2045 def : Pat<(v8f32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2046 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2047 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, VR256:$src2)),
2048 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2049 def : Pat<(v8i32 (X86Unpcklpsy VR256:$src1, (memopv8i32 addr:$src2))),
2050 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2051 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, (memopv8f32 addr:$src2))),
2052 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2053 def : Pat<(v8f32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2054 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2055 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, (memopv8i32 addr:$src2))),
2056 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2057 def : Pat<(v8i32 (X86Unpckhpsy VR256:$src1, VR256:$src2)),
2058 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2060 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, (memopv2f64 addr:$src2))),
2061 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2062 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1, VR128:$src2)),
2063 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2064 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, (memopv2f64 addr:$src2))),
2065 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2066 def : Pat<(v2f64 (X86Unpckhpd VR128:$src1, VR128:$src2)),
2067 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2069 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, (memopv4f64 addr:$src2))),
2070 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2071 def : Pat<(v4f64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2072 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2073 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, (memopv4i64 addr:$src2))),
2074 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2075 def : Pat<(v4i64 (X86Unpcklpdy VR256:$src1, VR256:$src2)),
2076 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2077 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, (memopv4f64 addr:$src2))),
2078 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2079 def : Pat<(v4f64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2080 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2081 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, (memopv4i64 addr:$src2))),
2082 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2083 def : Pat<(v4i64 (X86Unpckhpdy VR256:$src1, VR256:$src2)),
2084 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2086 // FIXME: Instead of X86Movddup, there should be a X86Unpcklpd here, the
2087 // problem is during lowering, where it's not possible to recognize the load
2088 // fold cause it has two uses through a bitcast. One use disappears at isel
2089 // time and the fold opportunity reappears.
2090 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2091 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2092 let AddedComplexity = 10 in
2093 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2094 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2097 //===----------------------------------------------------------------------===//
2098 // SSE 1 & 2 - Extract Floating-Point Sign mask
2099 //===----------------------------------------------------------------------===//
2101 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2102 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2104 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2105 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2106 [(set GR32:$dst, (Int RC:$src))], d>;
2107 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2108 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2111 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2112 SSEPackedSingle>, TB;
2113 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2114 SSEPackedDouble>, TB, OpSize;
2116 def : Pat<(i32 (X86fgetsign FR32:$src)),
2117 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2118 sub_ss))>, Requires<[HasSSE1]>;
2119 def : Pat<(i64 (X86fgetsign FR32:$src)),
2120 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2121 sub_ss))>, Requires<[HasSSE1]>;
2122 def : Pat<(i32 (X86fgetsign FR64:$src)),
2123 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2124 sub_sd))>, Requires<[HasSSE2]>;
2125 def : Pat<(i64 (X86fgetsign FR64:$src)),
2126 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2127 sub_sd))>, Requires<[HasSSE2]>;
2129 let Predicates = [HasAVX] in {
2130 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2131 "movmskps", SSEPackedSingle>, TB, VEX;
2132 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2133 "movmskpd", SSEPackedDouble>, TB, OpSize,
2135 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2136 "movmskps", SSEPackedSingle>, TB, VEX;
2137 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2138 "movmskpd", SSEPackedDouble>, TB, OpSize,
2141 def : Pat<(i32 (X86fgetsign FR32:$src)),
2142 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2144 def : Pat<(i64 (X86fgetsign FR32:$src)),
2145 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2147 def : Pat<(i32 (X86fgetsign FR64:$src)),
2148 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2150 def : Pat<(i64 (X86fgetsign FR64:$src)),
2151 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2155 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2156 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2157 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2158 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
2160 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2161 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2162 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2163 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB, OpSize,
2167 //===----------------------------------------------------------------------===//
2168 // SSE 1 & 2 - Misc aliasing of packed SSE 1 & 2 instructions
2169 //===----------------------------------------------------------------------===//
2171 // Aliases of packed SSE1 & SSE2 instructions for scalar use. These all have
2172 // names that start with 'Fs'.
2174 // Alias instructions that map fld0 to pxor for sse.
2175 let isReMaterializable = 1, isAsCheapAsAMove = 1, isCodeGenOnly = 1,
2176 canFoldAsLoad = 1 in {
2177 // FIXME: Set encoding to pseudo!
2178 def FsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2179 [(set FR32:$dst, fp32imm0)]>,
2180 Requires<[HasSSE1]>, TB, OpSize;
2181 def FsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2182 [(set FR64:$dst, fpimm0)]>,
2183 Requires<[HasSSE2]>, TB, OpSize;
2184 def VFsFLD0SS : I<0xEF, MRMInitReg, (outs FR32:$dst), (ins), "",
2185 [(set FR32:$dst, fp32imm0)]>,
2186 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2187 def VFsFLD0SD : I<0xEF, MRMInitReg, (outs FR64:$dst), (ins), "",
2188 [(set FR64:$dst, fpimm0)]>,
2189 Requires<[HasAVX]>, TB, OpSize, VEX_4V;
2192 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
2193 // bits are disregarded.
2194 let neverHasSideEffects = 1 in {
2195 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2196 "movaps\t{$src, $dst|$dst, $src}", []>;
2197 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2198 "movapd\t{$src, $dst|$dst, $src}", []>;
2201 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
2202 // bits are disregarded.
2203 let canFoldAsLoad = 1, isReMaterializable = 1 in {
2204 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
2205 "movaps\t{$src, $dst|$dst, $src}",
2206 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
2207 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
2208 "movapd\t{$src, $dst|$dst, $src}",
2209 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
2212 //===----------------------------------------------------------------------===//
2213 // SSE 1 & 2 - Logical Instructions
2214 //===----------------------------------------------------------------------===//
2216 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2218 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2220 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2221 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2223 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2224 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2226 let Constraints = "$src1 = $dst" in {
2227 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2228 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2230 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2231 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2235 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2236 let mayLoad = 0 in {
2237 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2238 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2239 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2242 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2243 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2245 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2247 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2249 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2250 // are all promoted to v2i64, and the patterns are covered by the int
2251 // version. This is needed in SSE only, because v2i64 isn't supported on
2252 // SSE1, but only on SSE2.
2253 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2254 !strconcat(OpcodeStr, "ps"), f128mem, [],
2255 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2256 (memopv2i64 addr:$src2)))], 0>, TB, VEX_4V;
2258 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2259 !strconcat(OpcodeStr, "pd"), f128mem,
2260 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2261 (bc_v2i64 (v2f64 VR128:$src2))))],
2262 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2263 (memopv2i64 addr:$src2)))], 0>,
2265 let Constraints = "$src1 = $dst" in {
2266 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2267 !strconcat(OpcodeStr, "ps"), f128mem,
2268 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2269 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2270 (memopv2i64 addr:$src2)))]>, TB;
2272 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2273 !strconcat(OpcodeStr, "pd"), f128mem,
2274 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2275 (bc_v2i64 (v2f64 VR128:$src2))))],
2276 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2277 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2281 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2283 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2285 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2286 !strconcat(OpcodeStr, "ps"), f256mem,
2287 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2288 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2289 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2291 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2292 !strconcat(OpcodeStr, "pd"), f256mem,
2293 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2294 (bc_v4i64 (v4f64 VR256:$src2))))],
2295 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2296 (memopv4i64 addr:$src2)))], 0>,
2300 // AVX 256-bit packed logical ops forms
2301 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2302 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2303 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2304 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2306 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2307 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2308 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2309 let isCommutable = 0 in
2310 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2312 //===----------------------------------------------------------------------===//
2313 // SSE 1 & 2 - Arithmetic Instructions
2314 //===----------------------------------------------------------------------===//
2316 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2319 /// In addition, we also have a special variant of the scalar form here to
2320 /// represent the associated intrinsic operation. This form is unlike the
2321 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2322 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2324 /// These three forms can each be reg+reg or reg+mem.
2327 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2329 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2331 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2332 OpNode, FR32, f32mem, Is2Addr>, XS;
2333 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2334 OpNode, FR64, f64mem, Is2Addr>, XD;
2337 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2339 let mayLoad = 0 in {
2340 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2341 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2342 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2343 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2347 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2349 let mayLoad = 0 in {
2350 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2351 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2352 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2353 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2357 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2359 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2360 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2361 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2362 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2365 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2367 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2368 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2369 SSEPackedSingle, Is2Addr>, TB;
2371 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2372 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2373 SSEPackedDouble, Is2Addr>, TB, OpSize;
2376 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2377 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2378 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2379 SSEPackedSingle, 0>, TB;
2381 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2382 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2383 SSEPackedDouble, 0>, TB, OpSize;
2386 // Binary Arithmetic instructions
2387 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2388 basic_sse12_fp_binop_s_int<0x58, "add", 0>,
2389 basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2390 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2391 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2392 basic_sse12_fp_binop_s_int<0x59, "mul", 0>,
2393 basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2394 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2396 let isCommutable = 0 in {
2397 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2398 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>,
2399 basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2400 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2401 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2402 basic_sse12_fp_binop_s_int<0x5E, "div", 0>,
2403 basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2404 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2405 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2406 basic_sse12_fp_binop_s_int<0x5F, "max", 0>,
2407 basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2408 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2409 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2410 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2411 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2412 basic_sse12_fp_binop_s_int<0x5D, "min", 0>,
2413 basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2414 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2415 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2416 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2419 let Constraints = "$src1 = $dst" in {
2420 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2421 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2422 basic_sse12_fp_binop_s_int<0x58, "add">;
2423 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2424 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2425 basic_sse12_fp_binop_s_int<0x59, "mul">;
2427 let isCommutable = 0 in {
2428 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2429 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2430 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2431 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2432 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2433 basic_sse12_fp_binop_s_int<0x5E, "div">;
2434 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2435 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2436 basic_sse12_fp_binop_s_int<0x5F, "max">,
2437 basic_sse12_fp_binop_p_int<0x5F, "max">;
2438 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2439 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2440 basic_sse12_fp_binop_s_int<0x5D, "min">,
2441 basic_sse12_fp_binop_p_int<0x5D, "min">;
2446 /// In addition, we also have a special variant of the scalar form here to
2447 /// represent the associated intrinsic operation. This form is unlike the
2448 /// plain scalar form, in that it takes an entire vector (instead of a
2449 /// scalar) and leaves the top elements undefined.
2451 /// And, we have a special variant form for a full-vector intrinsic form.
2453 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2454 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2455 SDNode OpNode, Intrinsic F32Int> {
2456 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2457 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2458 [(set FR32:$dst, (OpNode FR32:$src))]>;
2459 // For scalar unary operations, fold a load into the operation
2460 // only in OptForSize mode. It eliminates an instruction, but it also
2461 // eliminates a whole-register clobber (the load), so it introduces a
2462 // partial register update condition.
2463 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2464 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2465 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2466 Requires<[HasSSE1, OptForSize]>;
2467 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2468 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2469 [(set VR128:$dst, (F32Int VR128:$src))]>;
2470 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2471 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2472 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2475 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2476 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2477 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2478 !strconcat(OpcodeStr,
2479 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2480 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2481 !strconcat(OpcodeStr,
2482 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2483 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2484 (ins ssmem:$src1, VR128:$src2),
2485 !strconcat(OpcodeStr,
2486 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2489 /// sse1_fp_unop_p - SSE1 unops in packed form.
2490 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2491 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2492 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2493 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2494 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2495 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2496 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2499 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2500 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2501 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2502 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2503 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2504 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2505 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2506 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2509 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2510 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2511 Intrinsic V4F32Int> {
2512 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2513 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2514 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2515 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2516 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2517 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2520 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2521 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2522 Intrinsic V4F32Int> {
2523 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2524 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2525 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2526 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2527 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2528 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2531 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2532 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2533 SDNode OpNode, Intrinsic F64Int> {
2534 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2535 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2536 [(set FR64:$dst, (OpNode FR64:$src))]>;
2537 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2538 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2539 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2540 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2541 Requires<[HasSSE2, OptForSize]>;
2542 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2543 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2544 [(set VR128:$dst, (F64Int VR128:$src))]>;
2545 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2546 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2547 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2550 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2551 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2552 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2553 !strconcat(OpcodeStr,
2554 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2555 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2556 !strconcat(OpcodeStr,
2557 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2558 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2559 (ins VR128:$src1, sdmem:$src2),
2560 !strconcat(OpcodeStr,
2561 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2564 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2565 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2567 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2568 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2569 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2570 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2571 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2572 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2575 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2576 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2577 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2578 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2579 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2580 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2581 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2582 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2585 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2586 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2587 Intrinsic V2F64Int> {
2588 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2589 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2590 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2591 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2592 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2593 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2596 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2597 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2598 Intrinsic V2F64Int> {
2599 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2600 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2601 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2602 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2603 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2604 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2607 let Predicates = [HasAVX] in {
2609 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2610 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V;
2612 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2613 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2614 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2615 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2616 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2617 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2618 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2619 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2622 // Reciprocal approximations. Note that these typically require refinement
2623 // in order to obtain suitable precision.
2624 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V;
2625 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
2626 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
2627 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
2628 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
2630 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V;
2631 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
2632 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
2633 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
2634 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
2637 def : Pat<(f32 (fsqrt FR32:$src)),
2638 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2639 def : Pat<(f32 (fsqrt (load addr:$src))),
2640 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2641 Requires<[HasAVX, OptForSize]>;
2642 def : Pat<(f64 (fsqrt FR64:$src)),
2643 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
2644 def : Pat<(f64 (fsqrt (load addr:$src))),
2645 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
2646 Requires<[HasAVX, OptForSize]>;
2648 def : Pat<(f32 (X86frsqrt FR32:$src)),
2649 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2650 def : Pat<(f32 (X86frsqrt (load addr:$src))),
2651 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2652 Requires<[HasAVX, OptForSize]>;
2654 def : Pat<(f32 (X86frcp FR32:$src)),
2655 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
2656 def : Pat<(f32 (X86frcp (load addr:$src))),
2657 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
2658 Requires<[HasAVX, OptForSize]>;
2660 let Predicates = [HasAVX] in {
2661 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
2662 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2663 (VSQRTSSr (f32 (IMPLICIT_DEF)),
2664 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2666 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
2667 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2669 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
2670 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
2671 (VSQRTSDr (f64 (IMPLICIT_DEF)),
2672 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
2674 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
2675 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
2677 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
2678 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2679 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
2680 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2682 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
2683 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2685 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
2686 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
2687 (VRCPSSr (f32 (IMPLICIT_DEF)),
2688 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
2690 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
2691 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
2695 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
2696 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
2697 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
2698 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
2699 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
2700 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
2702 // Reciprocal approximations. Note that these typically require refinement
2703 // in order to obtain suitable precision.
2704 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
2705 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
2706 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
2707 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
2708 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
2709 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
2711 // There is no f64 version of the reciprocal approximation instructions.
2713 //===----------------------------------------------------------------------===//
2714 // SSE 1 & 2 - Non-temporal stores
2715 //===----------------------------------------------------------------------===//
2717 let AddedComplexity = 400 in { // Prefer non-temporal versions
2718 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
2719 (ins f128mem:$dst, VR128:$src),
2720 "movntps\t{$src, $dst|$dst, $src}",
2721 [(alignednontemporalstore (v4f32 VR128:$src),
2723 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
2724 (ins f128mem:$dst, VR128:$src),
2725 "movntpd\t{$src, $dst|$dst, $src}",
2726 [(alignednontemporalstore (v2f64 VR128:$src),
2728 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
2729 (ins f128mem:$dst, VR128:$src),
2730 "movntdq\t{$src, $dst|$dst, $src}",
2731 [(alignednontemporalstore (v2f64 VR128:$src),
2734 let ExeDomain = SSEPackedInt in
2735 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
2736 (ins f128mem:$dst, VR128:$src),
2737 "movntdq\t{$src, $dst|$dst, $src}",
2738 [(alignednontemporalstore (v4f32 VR128:$src),
2741 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2742 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
2744 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
2745 (ins f256mem:$dst, VR256:$src),
2746 "movntps\t{$src, $dst|$dst, $src}",
2747 [(alignednontemporalstore (v8f32 VR256:$src),
2749 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
2750 (ins f256mem:$dst, VR256:$src),
2751 "movntpd\t{$src, $dst|$dst, $src}",
2752 [(alignednontemporalstore (v4f64 VR256:$src),
2754 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
2755 (ins f256mem:$dst, VR256:$src),
2756 "movntdq\t{$src, $dst|$dst, $src}",
2757 [(alignednontemporalstore (v4f64 VR256:$src),
2759 let ExeDomain = SSEPackedInt in
2760 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
2761 (ins f256mem:$dst, VR256:$src),
2762 "movntdq\t{$src, $dst|$dst, $src}",
2763 [(alignednontemporalstore (v8f32 VR256:$src),
2767 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
2768 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
2769 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
2770 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
2771 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
2772 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
2774 let AddedComplexity = 400 in { // Prefer non-temporal versions
2775 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2776 "movntps\t{$src, $dst|$dst, $src}",
2777 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2778 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2779 "movntpd\t{$src, $dst|$dst, $src}",
2780 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
2782 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2783 "movntdq\t{$src, $dst|$dst, $src}",
2784 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
2786 let ExeDomain = SSEPackedInt in
2787 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
2788 "movntdq\t{$src, $dst|$dst, $src}",
2789 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
2791 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
2792 (MOVNTDQmr addr:$dst, VR128:$src)>;
2794 // There is no AVX form for instructions below this point
2795 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
2796 "movnti{l}\t{$src, $dst|$dst, $src}",
2797 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
2798 TB, Requires<[HasSSE2]>;
2799 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
2800 "movnti{q}\t{$src, $dst|$dst, $src}",
2801 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
2802 TB, Requires<[HasSSE2]>;
2805 //===----------------------------------------------------------------------===//
2806 // SSE 1 & 2 - Prefetch and memory fence
2807 //===----------------------------------------------------------------------===//
2809 // Prefetch intrinsic.
2810 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
2811 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
2812 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
2813 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
2814 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
2815 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
2816 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
2817 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
2819 // Load, store, and memory fence
2820 def SFENCE : I<0xAE, MRM_F8, (outs), (ins), "sfence", [(int_x86_sse_sfence)]>,
2821 TB, Requires<[HasSSE1]>;
2822 def : Pat<(X86SFence), (SFENCE)>;
2824 //===----------------------------------------------------------------------===//
2825 // SSE 1 & 2 - Load/Store XCSR register
2826 //===----------------------------------------------------------------------===//
2828 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2829 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
2830 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2831 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
2833 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
2834 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
2835 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
2836 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
2838 //===---------------------------------------------------------------------===//
2839 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
2840 //===---------------------------------------------------------------------===//
2842 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2844 let neverHasSideEffects = 1 in {
2845 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2846 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2847 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2848 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2850 def VMOVDQUrr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2851 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2852 def VMOVDQUYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2853 "movdqu\t{$src, $dst|$dst, $src}", []>, XS, VEX;
2855 let canFoldAsLoad = 1, mayLoad = 1 in {
2856 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2857 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2858 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2859 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2860 let Predicates = [HasAVX] in {
2861 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2862 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2863 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
2864 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2868 let mayStore = 1 in {
2869 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
2870 (ins i128mem:$dst, VR128:$src),
2871 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2872 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
2873 (ins i256mem:$dst, VR256:$src),
2874 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
2875 let Predicates = [HasAVX] in {
2876 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2877 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2878 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
2879 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
2883 let neverHasSideEffects = 1 in
2884 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2885 "movdqa\t{$src, $dst|$dst, $src}", []>;
2887 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2888 "movdqu\t{$src, $dst|$dst, $src}",
2889 []>, XS, Requires<[HasSSE2]>;
2891 let canFoldAsLoad = 1, mayLoad = 1 in {
2892 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2893 "movdqa\t{$src, $dst|$dst, $src}",
2894 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
2895 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
2896 "movdqu\t{$src, $dst|$dst, $src}",
2897 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
2898 XS, Requires<[HasSSE2]>;
2901 let mayStore = 1 in {
2902 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2903 "movdqa\t{$src, $dst|$dst, $src}",
2904 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
2905 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2906 "movdqu\t{$src, $dst|$dst, $src}",
2907 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
2908 XS, Requires<[HasSSE2]>;
2911 // Intrinsic forms of MOVDQU load and store
2912 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2913 "vmovdqu\t{$src, $dst|$dst, $src}",
2914 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2915 XS, VEX, Requires<[HasAVX]>;
2917 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
2918 "movdqu\t{$src, $dst|$dst, $src}",
2919 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
2920 XS, Requires<[HasSSE2]>;
2922 } // ExeDomain = SSEPackedInt
2924 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
2925 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
2926 (VMOVDQUYmr addr:$dst, VR256:$src)>;
2928 //===---------------------------------------------------------------------===//
2929 // SSE2 - Packed Integer Arithmetic Instructions
2930 //===---------------------------------------------------------------------===//
2932 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2934 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
2935 bit IsCommutable = 0, bit Is2Addr = 1> {
2936 let isCommutable = IsCommutable in
2937 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2938 (ins VR128:$src1, VR128:$src2),
2940 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2941 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2942 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2943 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2944 (ins VR128:$src1, i128mem:$src2),
2946 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2947 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2948 [(set VR128:$dst, (IntId VR128:$src1,
2949 (bitconvert (memopv2i64 addr:$src2))))]>;
2952 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
2953 string OpcodeStr, Intrinsic IntId,
2954 Intrinsic IntId2, bit Is2Addr = 1> {
2955 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2956 (ins VR128:$src1, VR128:$src2),
2958 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2959 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2960 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2))]>;
2961 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2962 (ins VR128:$src1, i128mem:$src2),
2964 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2965 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2966 [(set VR128:$dst, (IntId VR128:$src1,
2967 (bitconvert (memopv2i64 addr:$src2))))]>;
2968 def ri : PDIi8<opc2, ImmForm, (outs VR128:$dst),
2969 (ins VR128:$src1, i32i8imm:$src2),
2971 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2972 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2973 [(set VR128:$dst, (IntId2 VR128:$src1, (i32 imm:$src2)))]>;
2976 /// PDI_binop_rm - Simple SSE2 binary operator.
2977 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2978 ValueType OpVT, bit IsCommutable = 0, bit Is2Addr = 1> {
2979 let isCommutable = IsCommutable in
2980 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
2981 (ins VR128:$src1, VR128:$src2),
2983 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2984 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2985 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>;
2986 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
2987 (ins VR128:$src1, i128mem:$src2),
2989 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2990 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2991 [(set VR128:$dst, (OpVT (OpNode VR128:$src1,
2992 (bitconvert (memopv2i64 addr:$src2)))))]>;
2995 /// PDI_binop_rm_v2i64 - Simple SSE2 binary operator whose type is v2i64.
2997 /// FIXME: we could eliminate this and use PDI_binop_rm instead if tblgen knew
2998 /// to collapse (bitconvert VT to VT) into its operand.
3000 multiclass PDI_binop_rm_v2i64<bits<8> opc, string OpcodeStr, SDNode OpNode,
3001 bit IsCommutable = 0, bit Is2Addr = 1> {
3002 let isCommutable = IsCommutable in
3003 def rr : PDI<opc, MRMSrcReg, (outs VR128:$dst),
3004 (ins VR128:$src1, VR128:$src2),
3006 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3007 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3008 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))]>;
3009 def rm : PDI<opc, MRMSrcMem, (outs VR128:$dst),
3010 (ins VR128:$src1, i128mem:$src2),
3012 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3013 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3014 [(set VR128:$dst, (OpNode VR128:$src1, (memopv2i64 addr:$src2)))]>;
3017 } // ExeDomain = SSEPackedInt
3019 // 128-bit Integer Arithmetic
3021 let Predicates = [HasAVX] in {
3022 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, 1, 0 /*3addr*/>, VEX_4V;
3023 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, 1, 0>, VEX_4V;
3024 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, 1, 0>, VEX_4V;
3025 defm VPADDQ : PDI_binop_rm_v2i64<0xD4, "vpaddq", add, 1, 0>, VEX_4V;
3026 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, 1, 0>, VEX_4V;
3027 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, 0, 0>, VEX_4V;
3028 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, 0, 0>, VEX_4V;
3029 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, 0, 0>, VEX_4V;
3030 defm VPSUBQ : PDI_binop_rm_v2i64<0xFB, "vpsubq", sub, 0, 0>, VEX_4V;
3033 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b, 0, 0>,
3035 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w, 0, 0>,
3037 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b, 0, 0>,
3039 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w, 0, 0>,
3041 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b, 1, 0>,
3043 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w, 1, 0>,
3045 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b, 1, 0>,
3047 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w, 1, 0>,
3049 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w, 1, 0>,
3051 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w, 1, 0>,
3053 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq, 1, 0>,
3055 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd, 1, 0>,
3057 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b, 1, 0>,
3059 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w, 1, 0>,
3061 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b, 1, 0>,
3063 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w, 1, 0>,
3065 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b, 1, 0>,
3067 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w, 1, 0>,
3069 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw, 1, 0>,
3073 let Constraints = "$src1 = $dst" in {
3074 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, 1>;
3075 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, 1>;
3076 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, 1>;
3077 defm PADDQ : PDI_binop_rm_v2i64<0xD4, "paddq", add, 1>;
3078 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, 1>;
3079 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8>;
3080 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16>;
3081 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32>;
3082 defm PSUBQ : PDI_binop_rm_v2i64<0xFB, "psubq", sub>;
3085 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b>;
3086 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w>;
3087 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b>;
3088 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w>;
3089 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b, 1>;
3090 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w, 1>;
3091 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b, 1>;
3092 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w, 1>;
3093 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w, 1>;
3094 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w, 1>;
3095 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq, 1>;
3096 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd, 1>;
3097 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b, 1>;
3098 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w, 1>;
3099 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b, 1>;
3100 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w, 1>;
3101 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b, 1>;
3102 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w, 1>;
3103 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw, 1>;
3105 } // Constraints = "$src1 = $dst"
3107 //===---------------------------------------------------------------------===//
3108 // SSE2 - Packed Integer Logical Instructions
3109 //===---------------------------------------------------------------------===//
3111 let Predicates = [HasAVX] in {
3112 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3113 int_x86_sse2_psll_w, int_x86_sse2_pslli_w, 0>,
3115 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3116 int_x86_sse2_psll_d, int_x86_sse2_pslli_d, 0>,
3118 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3119 int_x86_sse2_psll_q, int_x86_sse2_pslli_q, 0>,
3122 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3123 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w, 0>,
3125 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3126 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d, 0>,
3128 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3129 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q, 0>,
3132 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3133 int_x86_sse2_psra_w, int_x86_sse2_psrai_w, 0>,
3135 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3136 int_x86_sse2_psra_d, int_x86_sse2_psrai_d, 0>,
3139 defm VPAND : PDI_binop_rm_v2i64<0xDB, "vpand", and, 1, 0>, VEX_4V;
3140 defm VPOR : PDI_binop_rm_v2i64<0xEB, "vpor" , or, 1, 0>, VEX_4V;
3141 defm VPXOR : PDI_binop_rm_v2i64<0xEF, "vpxor", xor, 1, 0>, VEX_4V;
3143 let ExeDomain = SSEPackedInt in {
3144 let neverHasSideEffects = 1 in {
3145 // 128-bit logical shifts.
3146 def VPSLLDQri : PDIi8<0x73, MRM7r,
3147 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3148 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3150 def VPSRLDQri : PDIi8<0x73, MRM3r,
3151 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3152 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3154 // PSRADQri doesn't exist in SSE[1-3].
3156 def VPANDNrr : PDI<0xDF, MRMSrcReg,
3157 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3158 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3160 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3162 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3163 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3164 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3165 [(set VR128:$dst, (X86andnp VR128:$src1,
3166 (memopv2i64 addr:$src2)))]>, VEX_4V;
3170 let Constraints = "$src1 = $dst" in {
3171 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3172 int_x86_sse2_psll_w, int_x86_sse2_pslli_w>;
3173 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3174 int_x86_sse2_psll_d, int_x86_sse2_pslli_d>;
3175 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3176 int_x86_sse2_psll_q, int_x86_sse2_pslli_q>;
3178 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3179 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w>;
3180 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3181 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d>;
3182 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3183 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q>;
3185 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3186 int_x86_sse2_psra_w, int_x86_sse2_psrai_w>;
3187 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3188 int_x86_sse2_psra_d, int_x86_sse2_psrai_d>;
3190 defm PAND : PDI_binop_rm_v2i64<0xDB, "pand", and, 1>;
3191 defm POR : PDI_binop_rm_v2i64<0xEB, "por" , or, 1>;
3192 defm PXOR : PDI_binop_rm_v2i64<0xEF, "pxor", xor, 1>;
3194 let ExeDomain = SSEPackedInt in {
3195 let neverHasSideEffects = 1 in {
3196 // 128-bit logical shifts.
3197 def PSLLDQri : PDIi8<0x73, MRM7r,
3198 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3199 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3200 def PSRLDQri : PDIi8<0x73, MRM3r,
3201 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3202 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3203 // PSRADQri doesn't exist in SSE[1-3].
3205 def PANDNrr : PDI<0xDF, MRMSrcReg,
3206 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3207 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3209 def PANDNrm : PDI<0xDF, MRMSrcMem,
3210 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3211 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3213 } // Constraints = "$src1 = $dst"
3215 let Predicates = [HasAVX] in {
3216 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3217 (v2i64 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3218 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3219 (v2i64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3220 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3221 (v2i64 (VPSLLDQri VR128:$src1, imm:$src2))>;
3222 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3223 (v2i64 (VPSRLDQri VR128:$src1, imm:$src2))>;
3224 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3225 (v2f64 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3227 // Shift up / down and insert zero's.
3228 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3229 (v2i64 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3230 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3231 (v2i64 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3234 let Predicates = [HasSSE2] in {
3235 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3236 (v2i64 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3237 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3238 (v2i64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3239 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3240 (v2i64 (PSLLDQri VR128:$src1, imm:$src2))>;
3241 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3242 (v2i64 (PSRLDQri VR128:$src1, imm:$src2))>;
3243 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3244 (v2f64 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2)))>;
3246 // Shift up / down and insert zero's.
3247 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3248 (v2i64 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3249 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3250 (v2i64 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt)))>;
3253 //===---------------------------------------------------------------------===//
3254 // SSE2 - Packed Integer Comparison Instructions
3255 //===---------------------------------------------------------------------===//
3257 let Predicates = [HasAVX] in {
3258 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b, 1,
3260 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w, 1,
3262 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d, 1,
3264 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b, 0,
3266 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w, 0,
3268 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d, 0,
3271 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3272 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3273 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3274 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3275 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3276 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3277 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3278 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3279 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3280 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3281 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3282 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3284 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3285 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3286 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3287 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3288 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3289 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3290 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3291 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3292 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3293 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3294 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3295 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3298 let Constraints = "$src1 = $dst" in {
3299 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b, 1>;
3300 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w, 1>;
3301 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d, 1>;
3302 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b>;
3303 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w>;
3304 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d>;
3305 } // Constraints = "$src1 = $dst"
3307 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3308 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3309 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, (memop addr:$src2))),
3310 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3311 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3312 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3313 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, (memop addr:$src2))),
3314 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3315 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3316 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3317 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, (memop addr:$src2))),
3318 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3320 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3321 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3322 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, (memop addr:$src2))),
3323 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3324 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3325 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3326 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, (memop addr:$src2))),
3327 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3328 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3329 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3330 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, (memop addr:$src2))),
3331 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3333 //===---------------------------------------------------------------------===//
3334 // SSE2 - Packed Integer Pack Instructions
3335 //===---------------------------------------------------------------------===//
3337 let Predicates = [HasAVX] in {
3338 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3340 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3342 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3346 let Constraints = "$src1 = $dst" in {
3347 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128>;
3348 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128>;
3349 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128>;
3350 } // Constraints = "$src1 = $dst"
3352 //===---------------------------------------------------------------------===//
3353 // SSE2 - Packed Integer Shuffle Instructions
3354 //===---------------------------------------------------------------------===//
3356 let ExeDomain = SSEPackedInt in {
3357 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
3359 def ri : Ii8<0x70, MRMSrcReg,
3360 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
3361 !strconcat(OpcodeStr,
3362 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3363 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
3365 def mi : Ii8<0x70, MRMSrcMem,
3366 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
3367 !strconcat(OpcodeStr,
3368 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
3369 [(set VR128:$dst, (vt (pshuf_frag:$src2
3370 (bc_frag (memopv2i64 addr:$src1)),
3373 } // ExeDomain = SSEPackedInt
3375 let Predicates = [HasAVX] in {
3376 let AddedComplexity = 5 in
3377 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
3380 // SSE2 with ImmT == Imm8 and XS prefix.
3381 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
3384 // SSE2 with ImmT == Imm8 and XD prefix.
3385 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
3388 let AddedComplexity = 5 in
3389 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3390 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3391 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
3392 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3393 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3395 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3397 (VPSHUFDmi addr:$src1, imm:$imm)>;
3398 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3400 (VPSHUFDmi addr:$src1, imm:$imm)>;
3401 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3402 (VPSHUFDri VR128:$src1, imm:$imm)>;
3403 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3404 (VPSHUFDri VR128:$src1, imm:$imm)>;
3405 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3406 (VPSHUFHWri VR128:$src, imm:$imm)>;
3407 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3409 (VPSHUFHWmi addr:$src, imm:$imm)>;
3410 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3411 (VPSHUFLWri VR128:$src, imm:$imm)>;
3412 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3414 (VPSHUFLWmi addr:$src, imm:$imm)>;
3417 let Predicates = [HasSSE2] in {
3418 let AddedComplexity = 5 in
3419 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
3421 // SSE2 with ImmT == Imm8 and XS prefix.
3422 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
3424 // SSE2 with ImmT == Imm8 and XD prefix.
3425 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
3427 let AddedComplexity = 5 in
3428 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
3429 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3430 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
3431 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
3432 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
3434 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
3436 (PSHUFDmi addr:$src1, imm:$imm)>;
3437 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
3439 (PSHUFDmi addr:$src1, imm:$imm)>;
3440 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3441 (PSHUFDri VR128:$src1, imm:$imm)>;
3442 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
3443 (PSHUFDri VR128:$src1, imm:$imm)>;
3444 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
3445 (PSHUFHWri VR128:$src, imm:$imm)>;
3446 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
3448 (PSHUFHWmi addr:$src, imm:$imm)>;
3449 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
3450 (PSHUFLWri VR128:$src, imm:$imm)>;
3451 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
3453 (PSHUFLWmi addr:$src, imm:$imm)>;
3456 //===---------------------------------------------------------------------===//
3457 // SSE2 - Packed Integer Unpack Instructions
3458 //===---------------------------------------------------------------------===//
3460 let ExeDomain = SSEPackedInt in {
3461 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
3462 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
3463 def rr : PDI<opc, MRMSrcReg,
3464 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3466 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3467 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3468 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
3469 def rm : PDI<opc, MRMSrcMem,
3470 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3472 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
3473 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3474 [(set VR128:$dst, (OpNode VR128:$src1,
3475 (bc_frag (memopv2i64
3479 let Predicates = [HasAVX] in {
3480 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Punpcklbw,
3481 bc_v16i8, 0>, VEX_4V;
3482 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Punpcklwd,
3483 bc_v8i16, 0>, VEX_4V;
3484 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Punpckldq,
3485 bc_v4i32, 0>, VEX_4V;
3487 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3488 /// knew to collapse (bitconvert VT to VT) into its operand.
3489 def VPUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3490 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3491 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3492 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3493 VR128:$src2)))]>, VEX_4V;
3494 def VPUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3495 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3496 "vpunpcklqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3497 [(set VR128:$dst, (v2i64 (X86Punpcklqdq VR128:$src1,
3498 (memopv2i64 addr:$src2))))]>, VEX_4V;
3500 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Punpckhbw,
3501 bc_v16i8, 0>, VEX_4V;
3502 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Punpckhwd,
3503 bc_v8i16, 0>, VEX_4V;
3504 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Punpckhdq,
3505 bc_v4i32, 0>, VEX_4V;
3507 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3508 /// knew to collapse (bitconvert VT to VT) into its operand.
3509 def VPUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3510 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3511 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3512 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3513 VR128:$src2)))]>, VEX_4V;
3514 def VPUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3515 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3516 "vpunpckhqdq\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3517 [(set VR128:$dst, (v2i64 (X86Punpckhqdq VR128:$src1,
3518 (memopv2i64 addr:$src2))))]>, VEX_4V;
3521 let Constraints = "$src1 = $dst" in {
3522 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Punpcklbw, bc_v16i8>;
3523 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Punpcklwd, bc_v8i16>;
3524 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Punpckldq, bc_v4i32>;
3526 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3527 /// knew to collapse (bitconvert VT to VT) into its operand.
3528 def PUNPCKLQDQrr : PDI<0x6C, MRMSrcReg,
3529 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3530 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3532 (v2i64 (X86Punpcklqdq VR128:$src1, VR128:$src2)))]>;
3533 def PUNPCKLQDQrm : PDI<0x6C, MRMSrcMem,
3534 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3535 "punpcklqdq\t{$src2, $dst|$dst, $src2}",
3537 (v2i64 (X86Punpcklqdq VR128:$src1,
3538 (memopv2i64 addr:$src2))))]>;
3540 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Punpckhbw, bc_v16i8>;
3541 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Punpckhwd, bc_v8i16>;
3542 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Punpckhdq, bc_v4i32>;
3544 /// FIXME: we could eliminate this and use sse2_unpack instead if tblgen
3545 /// knew to collapse (bitconvert VT to VT) into its operand.
3546 def PUNPCKHQDQrr : PDI<0x6D, MRMSrcReg,
3547 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3548 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3550 (v2i64 (X86Punpckhqdq VR128:$src1, VR128:$src2)))]>;
3551 def PUNPCKHQDQrm : PDI<0x6D, MRMSrcMem,
3552 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3553 "punpckhqdq\t{$src2, $dst|$dst, $src2}",
3555 (v2i64 (X86Punpckhqdq VR128:$src1,
3556 (memopv2i64 addr:$src2))))]>;
3559 } // ExeDomain = SSEPackedInt
3561 //===---------------------------------------------------------------------===//
3562 // SSE2 - Packed Integer Extract and Insert
3563 //===---------------------------------------------------------------------===//
3565 let ExeDomain = SSEPackedInt in {
3566 multiclass sse2_pinsrw<bit Is2Addr = 1> {
3567 def rri : Ii8<0xC4, MRMSrcReg,
3568 (outs VR128:$dst), (ins VR128:$src1,
3569 GR32:$src2, i32i8imm:$src3),
3571 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3572 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3574 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
3575 def rmi : Ii8<0xC4, MRMSrcMem,
3576 (outs VR128:$dst), (ins VR128:$src1,
3577 i16mem:$src2, i32i8imm:$src3),
3579 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
3580 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
3582 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
3587 let Predicates = [HasAVX] in
3588 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
3589 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3590 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3591 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3592 imm:$src2))]>, TB, OpSize, VEX;
3593 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
3594 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
3595 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3596 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
3600 let Predicates = [HasAVX] in {
3601 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
3602 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
3603 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
3604 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
3605 []>, TB, OpSize, VEX_4V;
3608 let Constraints = "$src1 = $dst" in
3609 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
3611 } // ExeDomain = SSEPackedInt
3613 //===---------------------------------------------------------------------===//
3614 // SSE2 - Packed Mask Creation
3615 //===---------------------------------------------------------------------===//
3617 let ExeDomain = SSEPackedInt in {
3619 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3620 "pmovmskb\t{$src, $dst|$dst, $src}",
3621 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
3622 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
3623 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
3624 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
3625 "pmovmskb\t{$src, $dst|$dst, $src}",
3626 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
3628 } // ExeDomain = SSEPackedInt
3630 //===---------------------------------------------------------------------===//
3631 // SSE2 - Conditional Store
3632 //===---------------------------------------------------------------------===//
3634 let ExeDomain = SSEPackedInt in {
3637 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
3638 (ins VR128:$src, VR128:$mask),
3639 "maskmovdqu\t{$mask, $src|$src, $mask}",
3640 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
3642 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
3643 (ins VR128:$src, VR128:$mask),
3644 "maskmovdqu\t{$mask, $src|$src, $mask}",
3645 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
3648 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3649 "maskmovdqu\t{$mask, $src|$src, $mask}",
3650 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
3652 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
3653 "maskmovdqu\t{$mask, $src|$src, $mask}",
3654 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
3656 } // ExeDomain = SSEPackedInt
3658 //===---------------------------------------------------------------------===//
3659 // SSE2 - Move Doubleword
3660 //===---------------------------------------------------------------------===//
3662 //===---------------------------------------------------------------------===//
3663 // Move Int Doubleword to Packed Double Int
3665 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3666 "movd\t{$src, $dst|$dst, $src}",
3668 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
3669 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3670 "movd\t{$src, $dst|$dst, $src}",
3672 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
3674 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3675 "mov{d|q}\t{$src, $dst|$dst, $src}",
3677 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
3678 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3679 "mov{d|q}\t{$src, $dst|$dst, $src}",
3680 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
3682 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3683 "movd\t{$src, $dst|$dst, $src}",
3685 (v4i32 (scalar_to_vector GR32:$src)))]>;
3686 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3687 "movd\t{$src, $dst|$dst, $src}",
3689 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
3690 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3691 "mov{d|q}\t{$src, $dst|$dst, $src}",
3693 (v2i64 (scalar_to_vector GR64:$src)))]>;
3694 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
3695 "mov{d|q}\t{$src, $dst|$dst, $src}",
3696 [(set FR64:$dst, (bitconvert GR64:$src))]>;
3698 //===---------------------------------------------------------------------===//
3699 // Move Int Doubleword to Single Scalar
3701 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3702 "movd\t{$src, $dst|$dst, $src}",
3703 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
3705 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3706 "movd\t{$src, $dst|$dst, $src}",
3707 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
3709 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
3710 "movd\t{$src, $dst|$dst, $src}",
3711 [(set FR32:$dst, (bitconvert GR32:$src))]>;
3713 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
3714 "movd\t{$src, $dst|$dst, $src}",
3715 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
3717 //===---------------------------------------------------------------------===//
3718 // Move Packed Doubleword Int to Packed Double Int
3720 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3721 "movd\t{$src, $dst|$dst, $src}",
3722 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3724 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
3725 (ins i32mem:$dst, VR128:$src),
3726 "movd\t{$src, $dst|$dst, $src}",
3727 [(store (i32 (vector_extract (v4i32 VR128:$src),
3728 (iPTR 0))), addr:$dst)]>, VEX;
3729 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
3730 "movd\t{$src, $dst|$dst, $src}",
3731 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
3733 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
3734 "movd\t{$src, $dst|$dst, $src}",
3735 [(store (i32 (vector_extract (v4i32 VR128:$src),
3736 (iPTR 0))), addr:$dst)]>;
3738 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3739 "mov{d|q}\t{$src, $dst|$dst, $src}",
3740 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
3742 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
3743 "movq\t{$src, $dst|$dst, $src}",
3744 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
3746 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
3747 "mov{d|q}\t{$src, $dst|$dst, $src}",
3748 [(set GR64:$dst, (bitconvert FR64:$src))]>;
3749 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
3750 "movq\t{$src, $dst|$dst, $src}",
3751 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
3753 //===---------------------------------------------------------------------===//
3754 // Move Scalar Single to Double Int
3756 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3757 "movd\t{$src, $dst|$dst, $src}",
3758 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
3759 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3760 "movd\t{$src, $dst|$dst, $src}",
3761 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
3762 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
3763 "movd\t{$src, $dst|$dst, $src}",
3764 [(set GR32:$dst, (bitconvert FR32:$src))]>;
3765 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
3766 "movd\t{$src, $dst|$dst, $src}",
3767 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
3769 //===---------------------------------------------------------------------===//
3770 // Patterns and instructions to describe movd/movq to XMM register zero-extends
3772 let AddedComplexity = 15 in {
3773 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3774 "movd\t{$src, $dst|$dst, $src}",
3775 [(set VR128:$dst, (v4i32 (X86vzmovl
3776 (v4i32 (scalar_to_vector GR32:$src)))))]>,
3778 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3779 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3780 [(set VR128:$dst, (v2i64 (X86vzmovl
3781 (v2i64 (scalar_to_vector GR64:$src)))))]>,
3784 let AddedComplexity = 15 in {
3785 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
3786 "movd\t{$src, $dst|$dst, $src}",
3787 [(set VR128:$dst, (v4i32 (X86vzmovl
3788 (v4i32 (scalar_to_vector GR32:$src)))))]>;
3789 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3790 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
3791 [(set VR128:$dst, (v2i64 (X86vzmovl
3792 (v2i64 (scalar_to_vector GR64:$src)))))]>;
3795 let AddedComplexity = 20 in {
3796 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3797 "movd\t{$src, $dst|$dst, $src}",
3799 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3800 (loadi32 addr:$src))))))]>,
3802 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
3803 "movd\t{$src, $dst|$dst, $src}",
3805 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
3806 (loadi32 addr:$src))))))]>;
3808 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
3809 (MOVZDI2PDIrm addr:$src)>;
3810 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
3811 (MOVZDI2PDIrm addr:$src)>;
3812 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
3813 (MOVZDI2PDIrm addr:$src)>;
3816 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
3817 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
3818 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
3819 (v4i32 (scalar_to_vector GR32:$src)), (i32 0)))),
3820 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
3821 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
3822 (v2i64 (scalar_to_vector GR64:$src)), (i32 0)))),
3823 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
3825 // These are the correct encodings of the instructions so that we know how to
3826 // read correct assembly, even though we continue to emit the wrong ones for
3827 // compatibility with Darwin's buggy assembler.
3828 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3829 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
3830 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3831 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
3832 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3833 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
3834 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3835 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
3836 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3837 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3838 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
3839 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
3841 //===---------------------------------------------------------------------===//
3842 // SSE2 - Move Quadword
3843 //===---------------------------------------------------------------------===//
3845 //===---------------------------------------------------------------------===//
3846 // Move Quadword Int to Packed Quadword Int
3848 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3849 "vmovq\t{$src, $dst|$dst, $src}",
3851 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3852 VEX, Requires<[HasAVX]>;
3853 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3854 "movq\t{$src, $dst|$dst, $src}",
3856 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
3857 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
3859 //===---------------------------------------------------------------------===//
3860 // Move Packed Quadword Int to Quadword Int
3862 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3863 "movq\t{$src, $dst|$dst, $src}",
3864 [(store (i64 (vector_extract (v2i64 VR128:$src),
3865 (iPTR 0))), addr:$dst)]>, VEX;
3866 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3867 "movq\t{$src, $dst|$dst, $src}",
3868 [(store (i64 (vector_extract (v2i64 VR128:$src),
3869 (iPTR 0))), addr:$dst)]>;
3871 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
3872 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
3874 //===---------------------------------------------------------------------===//
3875 // Store / copy lower 64-bits of a XMM register.
3877 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3878 "movq\t{$src, $dst|$dst, $src}",
3879 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
3880 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
3881 "movq\t{$src, $dst|$dst, $src}",
3882 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
3884 let AddedComplexity = 20 in
3885 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3886 "vmovq\t{$src, $dst|$dst, $src}",
3888 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3889 (loadi64 addr:$src))))))]>,
3890 XS, VEX, Requires<[HasAVX]>;
3892 let AddedComplexity = 20 in {
3893 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
3894 "movq\t{$src, $dst|$dst, $src}",
3896 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
3897 (loadi64 addr:$src))))))]>,
3898 XS, Requires<[HasSSE2]>;
3900 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
3901 (MOVZQI2PQIrm addr:$src)>;
3902 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
3903 (MOVZQI2PQIrm addr:$src)>;
3904 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
3907 //===---------------------------------------------------------------------===//
3908 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
3909 // IA32 document. movq xmm1, xmm2 does clear the high bits.
3911 let AddedComplexity = 15 in
3912 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3913 "vmovq\t{$src, $dst|$dst, $src}",
3914 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3915 XS, VEX, Requires<[HasAVX]>;
3916 let AddedComplexity = 15 in
3917 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3918 "movq\t{$src, $dst|$dst, $src}",
3919 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
3920 XS, Requires<[HasSSE2]>;
3922 let AddedComplexity = 20 in
3923 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3924 "vmovq\t{$src, $dst|$dst, $src}",
3925 [(set VR128:$dst, (v2i64 (X86vzmovl
3926 (loadv2i64 addr:$src))))]>,
3927 XS, VEX, Requires<[HasAVX]>;
3928 let AddedComplexity = 20 in {
3929 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3930 "movq\t{$src, $dst|$dst, $src}",
3931 [(set VR128:$dst, (v2i64 (X86vzmovl
3932 (loadv2i64 addr:$src))))]>,
3933 XS, Requires<[HasSSE2]>;
3935 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
3936 (MOVZPQILo2PQIrm addr:$src)>;
3939 // Instructions to match in the assembler
3940 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
3941 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3942 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3943 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3944 // Recognize "movd" with GR64 destination, but encode as a "movq"
3945 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
3946 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
3948 // Instructions for the disassembler
3949 // xr = XMM register
3952 let Predicates = [HasAVX] in
3953 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3954 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
3955 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3956 "movq\t{$src, $dst|$dst, $src}", []>, XS;
3958 //===---------------------------------------------------------------------===//
3959 // SSE2 - Misc Instructions
3960 //===---------------------------------------------------------------------===//
3963 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3964 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3965 TB, Requires<[HasSSE2]>;
3967 // Load, store, and memory fence
3968 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3969 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3970 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3971 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3972 def : Pat<(X86LFence), (LFENCE)>;
3973 def : Pat<(X86MFence), (MFENCE)>;
3976 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3977 // was introduced with SSE2, it's backward compatible.
3978 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3980 // Alias instructions that map zero vector to pxor / xorp* for sse.
3981 // We set canFoldAsLoad because this can be converted to a constant-pool
3982 // load of an all-ones value if folding it would be beneficial.
3983 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
3984 // JIT implementation, it does not expand the instructions below like
3985 // X86MCInstLower does.
3986 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3987 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in
3988 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3989 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
3990 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
3991 isCodeGenOnly = 1, ExeDomain = SSEPackedInt, Predicates = [HasAVX] in
3992 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
3993 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
3995 //===---------------------------------------------------------------------===//
3996 // SSE3 - Conversion Instructions
3997 //===---------------------------------------------------------------------===//
3999 // Convert Packed Double FP to Packed DW Integers
4000 let Predicates = [HasAVX] in {
4001 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4002 // register, but the same isn't true when using memory operands instead.
4003 // Provide other assembly rr and rm forms to address this explicitly.
4004 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4005 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4006 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4007 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4010 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4011 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4012 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4013 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4016 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4017 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4018 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4019 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4022 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4023 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4024 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4025 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4027 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4028 (VCVTPD2DQYrr VR256:$src)>;
4029 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4030 (VCVTPD2DQYrm addr:$src)>;
4032 // Convert Packed DW Integers to Packed Double FP
4033 let Predicates = [HasAVX] in {
4034 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4035 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4036 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4037 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4038 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4039 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4040 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4041 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4044 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4045 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4046 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4047 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4049 // AVX 256-bit register conversion intrinsics
4050 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4051 (VCVTDQ2PDYrr VR128:$src)>;
4052 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
4053 (VCVTDQ2PDYrm addr:$src)>;
4055 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4056 (VCVTPD2DQYrr VR256:$src)>;
4057 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4058 (VCVTPD2DQYrm addr:$src)>;
4060 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4061 (VCVTDQ2PDYrr VR128:$src)>;
4062 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
4063 (VCVTDQ2PDYrm addr:$src)>;
4065 //===---------------------------------------------------------------------===//
4066 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4067 //===---------------------------------------------------------------------===//
4068 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4069 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4070 X86MemOperand x86memop> {
4071 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4072 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4073 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4074 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4075 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4076 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4079 let Predicates = [HasAVX] in {
4080 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4081 v4f32, VR128, memopv4f32, f128mem>, VEX;
4082 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4083 v4f32, VR128, memopv4f32, f128mem>, VEX;
4084 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4085 v8f32, VR256, memopv8f32, f256mem>, VEX;
4086 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4087 v8f32, VR256, memopv8f32, f256mem>, VEX;
4089 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4090 memopv4f32, f128mem>;
4091 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4092 memopv4f32, f128mem>;
4094 let Predicates = [HasSSE3] in {
4095 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4096 (MOVSHDUPrr VR128:$src)>;
4097 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4098 (MOVSHDUPrm addr:$src)>;
4099 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4100 (MOVSLDUPrr VR128:$src)>;
4101 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4102 (MOVSLDUPrm addr:$src)>;
4105 let Predicates = [HasAVX] in {
4106 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4107 (VMOVSHDUPrr VR128:$src)>;
4108 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4109 (VMOVSHDUPrm addr:$src)>;
4110 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4111 (VMOVSLDUPrr VR128:$src)>;
4112 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4113 (VMOVSLDUPrm addr:$src)>;
4114 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4115 (VMOVSHDUPYrr VR256:$src)>;
4116 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4117 (VMOVSHDUPYrm addr:$src)>;
4118 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4119 (VMOVSLDUPYrr VR256:$src)>;
4120 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4121 (VMOVSLDUPYrm addr:$src)>;
4124 //===---------------------------------------------------------------------===//
4125 // SSE3 - Replicate Double FP - MOVDDUP
4126 //===---------------------------------------------------------------------===//
4128 multiclass sse3_replicate_dfp<string OpcodeStr> {
4129 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4130 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4131 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4132 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4133 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4135 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4139 // FIXME: Merge with above classe when there're patterns for the ymm version
4140 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4141 let Predicates = [HasAVX] in {
4142 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4143 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4145 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4146 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4151 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4152 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4153 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4155 let Predicates = [HasSSE3] in {
4156 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4158 (MOVDDUPrm addr:$src)>;
4159 let AddedComplexity = 5 in {
4160 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4161 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4162 (MOVDDUPrm addr:$src)>;
4163 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4164 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4165 (MOVDDUPrm addr:$src)>;
4167 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4168 (MOVDDUPrm addr:$src)>;
4169 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4170 (MOVDDUPrm addr:$src)>;
4171 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4172 (MOVDDUPrm addr:$src)>;
4173 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4174 (MOVDDUPrm addr:$src)>;
4175 def : Pat<(X86Movddup (bc_v2f64
4176 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4177 (MOVDDUPrm addr:$src)>;
4180 let Predicates = [HasAVX] in {
4181 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4183 (VMOVDDUPrm addr:$src)>;
4184 let AddedComplexity = 5 in {
4185 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4186 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4187 (VMOVDDUPrm addr:$src)>;
4188 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4189 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4190 (VMOVDDUPrm addr:$src)>;
4192 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4193 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4194 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4195 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4196 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4197 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4198 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4199 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4200 def : Pat<(X86Movddup (bc_v2f64
4201 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4202 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4205 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4206 (VMOVDDUPYrm addr:$src)>;
4207 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4208 (VMOVDDUPYrm addr:$src)>;
4209 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4210 (VMOVDDUPYrm addr:$src)>;
4211 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4212 (VMOVDDUPYrm addr:$src)>;
4213 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4214 (VMOVDDUPYrr VR256:$src)>;
4215 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4216 (VMOVDDUPYrr VR256:$src)>;
4219 //===---------------------------------------------------------------------===//
4220 // SSE3 - Move Unaligned Integer
4221 //===---------------------------------------------------------------------===//
4223 let Predicates = [HasAVX] in {
4224 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4225 "vlddqu\t{$src, $dst|$dst, $src}",
4226 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4227 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4228 "vlddqu\t{$src, $dst|$dst, $src}",
4229 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4231 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4232 "lddqu\t{$src, $dst|$dst, $src}",
4233 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4235 //===---------------------------------------------------------------------===//
4236 // SSE3 - Arithmetic
4237 //===---------------------------------------------------------------------===//
4239 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4240 X86MemOperand x86memop, bit Is2Addr = 1> {
4241 def rr : I<0xD0, MRMSrcReg,
4242 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4244 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4245 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4246 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4247 def rm : I<0xD0, MRMSrcMem,
4248 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4250 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4251 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4252 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4255 let Predicates = [HasAVX],
4256 ExeDomain = SSEPackedDouble in {
4257 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
4258 f128mem, 0>, TB, XD, VEX_4V;
4259 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
4260 f128mem, 0>, TB, OpSize, VEX_4V;
4261 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
4262 f256mem, 0>, TB, XD, VEX_4V;
4263 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
4264 f256mem, 0>, TB, OpSize, VEX_4V;
4266 let Constraints = "$src1 = $dst", Predicates = [HasSSE3],
4267 ExeDomain = SSEPackedDouble in {
4268 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
4270 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
4271 f128mem>, TB, OpSize;
4274 //===---------------------------------------------------------------------===//
4275 // SSE3 Instructions
4276 //===---------------------------------------------------------------------===//
4279 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4280 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4281 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4283 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4284 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4285 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4287 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4289 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4290 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4291 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4293 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
4294 X86MemOperand x86memop, Intrinsic IntId, bit Is2Addr = 1> {
4295 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
4297 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4298 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4299 [(set RC:$dst, (vt (IntId RC:$src1, RC:$src2)))]>;
4301 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4303 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4304 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4305 [(set RC:$dst, (vt (IntId RC:$src1, (memop addr:$src2))))]>;
4308 let Predicates = [HasAVX] in {
4309 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
4310 int_x86_sse3_hadd_ps, 0>, VEX_4V;
4311 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
4312 int_x86_sse3_hadd_pd, 0>, VEX_4V;
4313 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
4314 int_x86_sse3_hsub_ps, 0>, VEX_4V;
4315 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
4316 int_x86_sse3_hsub_pd, 0>, VEX_4V;
4317 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
4318 int_x86_avx_hadd_ps_256, 0>, VEX_4V;
4319 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
4320 int_x86_avx_hadd_pd_256, 0>, VEX_4V;
4321 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
4322 int_x86_avx_hsub_ps_256, 0>, VEX_4V;
4323 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
4324 int_x86_avx_hsub_pd_256, 0>, VEX_4V;
4327 let Constraints = "$src1 = $dst" in {
4328 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem,
4329 int_x86_sse3_hadd_ps>;
4330 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem,
4331 int_x86_sse3_hadd_pd>;
4332 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem,
4333 int_x86_sse3_hsub_ps>;
4334 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem,
4335 int_x86_sse3_hsub_pd>;
4338 //===---------------------------------------------------------------------===//
4339 // SSSE3 - Packed Absolute Instructions
4340 //===---------------------------------------------------------------------===//
4343 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
4344 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
4345 PatFrag mem_frag128, Intrinsic IntId128> {
4346 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4348 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4349 [(set VR128:$dst, (IntId128 VR128:$src))]>,
4352 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4354 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4357 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
4360 let Predicates = [HasAVX] in {
4361 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
4362 int_x86_ssse3_pabs_b_128>, VEX;
4363 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
4364 int_x86_ssse3_pabs_w_128>, VEX;
4365 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
4366 int_x86_ssse3_pabs_d_128>, VEX;
4369 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
4370 int_x86_ssse3_pabs_b_128>;
4371 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
4372 int_x86_ssse3_pabs_w_128>;
4373 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
4374 int_x86_ssse3_pabs_d_128>;
4376 //===---------------------------------------------------------------------===//
4377 // SSSE3 - Packed Binary Operator Instructions
4378 //===---------------------------------------------------------------------===//
4380 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
4381 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
4382 PatFrag mem_frag128, Intrinsic IntId128,
4384 let isCommutable = 1 in
4385 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
4386 (ins VR128:$src1, VR128:$src2),
4388 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4389 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4390 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
4392 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
4393 (ins VR128:$src1, i128mem:$src2),
4395 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4396 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4398 (IntId128 VR128:$src1,
4399 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
4402 let Predicates = [HasAVX] in {
4403 let isCommutable = 0 in {
4404 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
4405 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
4406 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
4407 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
4408 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
4409 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
4410 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
4411 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
4412 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
4413 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
4414 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
4415 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
4416 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
4417 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
4418 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
4419 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
4420 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
4421 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
4422 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
4423 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
4424 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
4425 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
4427 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
4428 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
4431 // None of these have i8 immediate fields.
4432 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
4433 let isCommutable = 0 in {
4434 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
4435 int_x86_ssse3_phadd_w_128>;
4436 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
4437 int_x86_ssse3_phadd_d_128>;
4438 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
4439 int_x86_ssse3_phadd_sw_128>;
4440 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
4441 int_x86_ssse3_phsub_w_128>;
4442 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
4443 int_x86_ssse3_phsub_d_128>;
4444 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
4445 int_x86_ssse3_phsub_sw_128>;
4446 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
4447 int_x86_ssse3_pmadd_ub_sw_128>;
4448 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
4449 int_x86_ssse3_pshuf_b_128>;
4450 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
4451 int_x86_ssse3_psign_b_128>;
4452 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
4453 int_x86_ssse3_psign_w_128>;
4454 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
4455 int_x86_ssse3_psign_d_128>;
4457 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
4458 int_x86_ssse3_pmul_hr_sw_128>;
4461 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
4462 (PSHUFBrr128 VR128:$src, VR128:$mask)>, Requires<[HasSSSE3]>;
4463 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
4464 (PSHUFBrm128 VR128:$src, addr:$mask)>, Requires<[HasSSSE3]>;
4466 def : Pat<(X86psignb VR128:$src1, VR128:$src2),
4467 (PSIGNBrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4468 def : Pat<(X86psignw VR128:$src1, VR128:$src2),
4469 (PSIGNWrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4470 def : Pat<(X86psignd VR128:$src1, VR128:$src2),
4471 (PSIGNDrr128 VR128:$src1, VR128:$src2)>, Requires<[HasSSSE3]>;
4473 //===---------------------------------------------------------------------===//
4474 // SSSE3 - Packed Align Instruction Patterns
4475 //===---------------------------------------------------------------------===//
4477 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
4478 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
4479 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
4481 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4483 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4485 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
4486 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
4488 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4490 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4494 let Predicates = [HasAVX] in
4495 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
4496 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
4497 defm PALIGN : ssse3_palign<"palignr">;
4499 let Predicates = [HasSSSE3] in {
4500 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4501 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4502 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4503 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4504 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4505 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4506 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4507 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4510 let Predicates = [HasAVX] in {
4511 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4512 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4513 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4514 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4515 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4516 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4517 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
4518 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
4521 //===---------------------------------------------------------------------===//
4522 // SSSE3 Misc Instructions
4523 //===---------------------------------------------------------------------===//
4525 // Thread synchronization
4526 let usesCustomInserter = 1 in {
4527 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
4528 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
4529 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
4530 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
4533 let Uses = [EAX, ECX, EDX] in
4534 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
4535 Requires<[HasSSE3]>;
4536 let Uses = [ECX, EAX] in
4537 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
4538 Requires<[HasSSE3]>;
4540 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
4541 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
4543 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
4544 Requires<[In32BitMode]>;
4545 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
4546 Requires<[In64BitMode]>;
4548 // extload f32 -> f64. This matches load+fextend because we have a hack in
4549 // the isel (PreprocessForFPConvert) that can introduce loads after dag
4551 // Since these loads aren't folded into the fextend, we have to match it
4553 let Predicates = [HasSSE2] in
4554 def : Pat<(fextend (loadf32 addr:$src)),
4555 (CVTSS2SDrm addr:$src)>;
4557 // Splat v2f64 / v2i64
4558 let AddedComplexity = 10 in {
4559 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4560 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4563 let AddedComplexity = 20 in {
4564 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
4565 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
4566 (MOVLPSrm VR128:$src1, addr:$src2)>;
4567 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
4568 (MOVLPDrm VR128:$src1, addr:$src2)>;
4569 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
4570 (MOVLPSrm VR128:$src1, addr:$src2)>;
4571 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
4572 (MOVLPDrm VR128:$src1, addr:$src2)>;
4575 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
4576 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4577 (MOVLPSmr addr:$src1, VR128:$src2)>;
4578 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4579 (MOVLPDmr addr:$src1, VR128:$src2)>;
4580 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
4582 (MOVLPSmr addr:$src1, VR128:$src2)>;
4583 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
4584 (MOVLPDmr addr:$src1, VR128:$src2)>;
4586 // Set lowest element and zero upper elements.
4587 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4588 (MOVZPQILo2PQIrr VR128:$src)>, Requires<[HasSSE2]>;
4590 // Use movaps / movups for SSE integer load / store (one byte shorter).
4591 // The instructions selected below are then converted to MOVDQA/MOVDQU
4592 // during the SSE domain pass.
4593 let Predicates = [HasSSE1] in {
4594 def : Pat<(alignedloadv4i32 addr:$src),
4595 (MOVAPSrm addr:$src)>;
4596 def : Pat<(loadv4i32 addr:$src),
4597 (MOVUPSrm addr:$src)>;
4598 def : Pat<(alignedloadv2i64 addr:$src),
4599 (MOVAPSrm addr:$src)>;
4600 def : Pat<(loadv2i64 addr:$src),
4601 (MOVUPSrm addr:$src)>;
4603 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4604 (MOVAPSmr addr:$dst, VR128:$src)>;
4605 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4606 (MOVAPSmr addr:$dst, VR128:$src)>;
4607 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4608 (MOVAPSmr addr:$dst, VR128:$src)>;
4609 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4610 (MOVAPSmr addr:$dst, VR128:$src)>;
4611 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4612 (MOVUPSmr addr:$dst, VR128:$src)>;
4613 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4614 (MOVUPSmr addr:$dst, VR128:$src)>;
4615 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4616 (MOVUPSmr addr:$dst, VR128:$src)>;
4617 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4618 (MOVUPSmr addr:$dst, VR128:$src)>;
4621 // Use vmovaps/vmovups for AVX integer load/store.
4622 let Predicates = [HasAVX] in {
4623 // 128-bit load/store
4624 def : Pat<(alignedloadv4i32 addr:$src),
4625 (VMOVAPSrm addr:$src)>;
4626 def : Pat<(loadv4i32 addr:$src),
4627 (VMOVUPSrm addr:$src)>;
4628 def : Pat<(alignedloadv2i64 addr:$src),
4629 (VMOVAPSrm addr:$src)>;
4630 def : Pat<(loadv2i64 addr:$src),
4631 (VMOVUPSrm addr:$src)>;
4633 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
4634 (VMOVAPSmr addr:$dst, VR128:$src)>;
4635 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
4636 (VMOVAPSmr addr:$dst, VR128:$src)>;
4637 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
4638 (VMOVAPSmr addr:$dst, VR128:$src)>;
4639 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
4640 (VMOVAPSmr addr:$dst, VR128:$src)>;
4641 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
4642 (VMOVUPSmr addr:$dst, VR128:$src)>;
4643 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
4644 (VMOVUPSmr addr:$dst, VR128:$src)>;
4645 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
4646 (VMOVUPSmr addr:$dst, VR128:$src)>;
4647 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
4648 (VMOVUPSmr addr:$dst, VR128:$src)>;
4650 // 256-bit load/store
4651 def : Pat<(alignedloadv4i64 addr:$src),
4652 (VMOVAPSYrm addr:$src)>;
4653 def : Pat<(loadv4i64 addr:$src),
4654 (VMOVUPSYrm addr:$src)>;
4655 def : Pat<(alignedloadv8i32 addr:$src),
4656 (VMOVAPSYrm addr:$src)>;
4657 def : Pat<(loadv8i32 addr:$src),
4658 (VMOVUPSYrm addr:$src)>;
4659 def : Pat<(alignedstore (v4i64 VR256:$src), addr:$dst),
4660 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4661 def : Pat<(alignedstore (v8i32 VR256:$src), addr:$dst),
4662 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4663 def : Pat<(alignedstore (v16i16 VR256:$src), addr:$dst),
4664 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4665 def : Pat<(alignedstore (v32i8 VR256:$src), addr:$dst),
4666 (VMOVAPSYmr addr:$dst, VR256:$src)>;
4667 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
4668 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4669 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
4670 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4671 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
4672 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4673 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
4674 (VMOVUPSYmr addr:$dst, VR256:$src)>;
4677 //===----------------------------------------------------------------------===//
4678 // SSE4.1 - Packed Move with Sign/Zero Extend
4679 //===----------------------------------------------------------------------===//
4681 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4682 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4683 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4684 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4686 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4687 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4689 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
4693 let Predicates = [HasAVX] in {
4694 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
4696 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
4698 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
4700 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
4702 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
4704 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
4708 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
4709 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
4710 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
4711 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
4712 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
4713 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
4715 // Common patterns involving scalar load.
4716 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
4717 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4718 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
4719 (PMOVSXBWrm addr:$src)>, Requires<[HasSSE41]>;
4721 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
4722 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4723 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
4724 (PMOVSXWDrm addr:$src)>, Requires<[HasSSE41]>;
4726 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
4727 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4728 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
4729 (PMOVSXDQrm addr:$src)>, Requires<[HasSSE41]>;
4731 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
4732 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4733 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
4734 (PMOVZXBWrm addr:$src)>, Requires<[HasSSE41]>;
4736 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
4737 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4738 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
4739 (PMOVZXWDrm addr:$src)>, Requires<[HasSSE41]>;
4741 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
4742 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4743 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
4744 (PMOVZXDQrm addr:$src)>, Requires<[HasSSE41]>;
4747 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4748 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4749 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4750 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4752 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4753 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4755 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
4759 let Predicates = [HasAVX] in {
4760 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
4762 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
4764 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
4766 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
4770 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
4771 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
4772 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
4773 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
4775 // Common patterns involving scalar load
4776 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
4777 (PMOVSXBDrm addr:$src)>, Requires<[HasSSE41]>;
4778 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
4779 (PMOVSXWQrm addr:$src)>, Requires<[HasSSE41]>;
4781 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
4782 (PMOVZXBDrm addr:$src)>, Requires<[HasSSE41]>;
4783 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
4784 (PMOVZXWQrm addr:$src)>, Requires<[HasSSE41]>;
4787 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
4788 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4789 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4790 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
4792 // Expecting a i16 load any extended to i32 value.
4793 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
4794 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4795 [(set VR128:$dst, (IntId (bitconvert
4796 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
4800 let Predicates = [HasAVX] in {
4801 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
4803 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
4806 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
4807 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
4809 // Common patterns involving scalar load
4810 def : Pat<(int_x86_sse41_pmovsxbq
4811 (bitconvert (v4i32 (X86vzmovl
4812 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4813 (PMOVSXBQrm addr:$src)>, Requires<[HasSSE41]>;
4815 def : Pat<(int_x86_sse41_pmovzxbq
4816 (bitconvert (v4i32 (X86vzmovl
4817 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
4818 (PMOVZXBQrm addr:$src)>, Requires<[HasSSE41]>;
4820 //===----------------------------------------------------------------------===//
4821 // SSE4.1 - Extract Instructions
4822 //===----------------------------------------------------------------------===//
4824 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
4825 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
4826 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4827 (ins VR128:$src1, i32i8imm:$src2),
4828 !strconcat(OpcodeStr,
4829 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4830 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
4832 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4833 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
4834 !strconcat(OpcodeStr,
4835 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4838 // There's an AssertZext in the way of writing the store pattern
4839 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4842 let Predicates = [HasAVX] in {
4843 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
4844 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
4845 (ins VR128:$src1, i32i8imm:$src2),
4846 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
4849 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
4852 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
4853 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
4854 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4855 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
4856 !strconcat(OpcodeStr,
4857 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4860 // There's an AssertZext in the way of writing the store pattern
4861 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
4864 let Predicates = [HasAVX] in
4865 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
4867 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
4870 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4871 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
4872 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4873 (ins VR128:$src1, i32i8imm:$src2),
4874 !strconcat(OpcodeStr,
4875 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4877 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
4878 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4879 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
4880 !strconcat(OpcodeStr,
4881 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4882 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
4883 addr:$dst)]>, OpSize;
4886 let Predicates = [HasAVX] in
4887 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
4889 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
4891 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
4892 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
4893 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
4894 (ins VR128:$src1, i32i8imm:$src2),
4895 !strconcat(OpcodeStr,
4896 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4898 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
4899 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4900 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
4901 !strconcat(OpcodeStr,
4902 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4903 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
4904 addr:$dst)]>, OpSize, REX_W;
4907 let Predicates = [HasAVX] in
4908 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
4910 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
4912 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
4914 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
4915 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
4916 (ins VR128:$src1, i32i8imm:$src2),
4917 !strconcat(OpcodeStr,
4918 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4920 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
4922 def mr : SS4AIi8<opc, MRMDestMem, (outs),
4923 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
4924 !strconcat(OpcodeStr,
4925 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4926 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
4927 addr:$dst)]>, OpSize;
4930 let Predicates = [HasAVX] in {
4931 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
4932 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
4933 (ins VR128:$src1, i32i8imm:$src2),
4934 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
4937 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
4939 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
4940 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
4943 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
4944 Requires<[HasSSE41]>;
4946 //===----------------------------------------------------------------------===//
4947 // SSE4.1 - Insert Instructions
4948 //===----------------------------------------------------------------------===//
4950 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
4951 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4952 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4954 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4956 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4958 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
4959 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4960 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
4962 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4964 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4966 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
4967 imm:$src3))]>, OpSize;
4970 let Predicates = [HasAVX] in
4971 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
4972 let Constraints = "$src1 = $dst" in
4973 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
4975 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
4976 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
4977 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
4979 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4981 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4983 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
4985 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
4986 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
4988 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
4990 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
4992 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
4993 imm:$src3)))]>, OpSize;
4996 let Predicates = [HasAVX] in
4997 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
4998 let Constraints = "$src1 = $dst" in
4999 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5001 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5002 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5003 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5005 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5007 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5009 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5011 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5012 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5014 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5016 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5018 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5019 imm:$src3)))]>, OpSize;
5022 let Predicates = [HasAVX] in
5023 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5024 let Constraints = "$src1 = $dst" in
5025 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5027 // insertps has a few different modes, there's the first two here below which
5028 // are optimized inserts that won't zero arbitrary elements in the destination
5029 // vector. The next one matches the intrinsic and could zero arbitrary elements
5030 // in the target vector.
5031 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5032 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5033 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5035 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5037 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5039 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5041 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5042 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5044 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5046 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5048 (X86insrtps VR128:$src1,
5049 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5050 imm:$src3))]>, OpSize;
5053 let Constraints = "$src1 = $dst" in
5054 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5055 let Predicates = [HasAVX] in
5056 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5058 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5059 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5061 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5062 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5063 Requires<[HasSSE41]>;
5065 //===----------------------------------------------------------------------===//
5066 // SSE4.1 - Round Instructions
5067 //===----------------------------------------------------------------------===//
5069 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5070 X86MemOperand x86memop, RegisterClass RC,
5071 PatFrag mem_frag32, PatFrag mem_frag64,
5072 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5073 // Intrinsic operation, reg.
5074 // Vector intrinsic operation, reg
5075 def PSr : SS4AIi8<opcps, MRMSrcReg,
5076 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5077 !strconcat(OpcodeStr,
5078 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5079 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5082 // Vector intrinsic operation, mem
5083 def PSm : Ii8<opcps, MRMSrcMem,
5084 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
5085 !strconcat(OpcodeStr,
5086 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5088 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
5090 Requires<[HasSSE41]>;
5092 // Vector intrinsic operation, reg
5093 def PDr : SS4AIi8<opcpd, MRMSrcReg,
5094 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5095 !strconcat(OpcodeStr,
5096 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5097 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
5100 // Vector intrinsic operation, mem
5101 def PDm : SS4AIi8<opcpd, MRMSrcMem,
5102 (outs RC:$dst), (ins f256mem:$src1, i32i8imm:$src2),
5103 !strconcat(OpcodeStr,
5104 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5106 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
5110 multiclass sse41_fp_unop_rm_avx_p<bits<8> opcps, bits<8> opcpd,
5111 RegisterClass RC, X86MemOperand x86memop, string OpcodeStr> {
5112 // Intrinsic operation, reg.
5113 // Vector intrinsic operation, reg
5114 def PSr_AVX : SS4AIi8<opcps, MRMSrcReg,
5115 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5116 !strconcat(OpcodeStr,
5117 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5120 // Vector intrinsic operation, mem
5121 def PSm_AVX : Ii8<opcps, MRMSrcMem,
5122 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5123 !strconcat(OpcodeStr,
5124 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5125 []>, TA, OpSize, Requires<[HasSSE41]>;
5127 // Vector intrinsic operation, reg
5128 def PDr_AVX : SS4AIi8<opcpd, MRMSrcReg,
5129 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5130 !strconcat(OpcodeStr,
5131 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5134 // Vector intrinsic operation, mem
5135 def PDm_AVX : SS4AIi8<opcpd, MRMSrcMem,
5136 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5137 !strconcat(OpcodeStr,
5138 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5142 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
5145 Intrinsic F64Int, bit Is2Addr = 1> {
5146 // Intrinsic operation, reg.
5147 def SSr : SS4AIi8<opcss, MRMSrcReg,
5148 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5150 !strconcat(OpcodeStr,
5151 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5152 !strconcat(OpcodeStr,
5153 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5154 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5157 // Intrinsic operation, mem.
5158 def SSm : SS4AIi8<opcss, MRMSrcMem,
5159 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5161 !strconcat(OpcodeStr,
5162 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5163 !strconcat(OpcodeStr,
5164 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5166 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
5169 // Intrinsic operation, reg.
5170 def SDr : SS4AIi8<opcsd, MRMSrcReg,
5171 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5173 !strconcat(OpcodeStr,
5174 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5175 !strconcat(OpcodeStr,
5176 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5177 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
5180 // Intrinsic operation, mem.
5181 def SDm : SS4AIi8<opcsd, MRMSrcMem,
5182 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5184 !strconcat(OpcodeStr,
5185 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5186 !strconcat(OpcodeStr,
5187 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5189 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
5193 multiclass sse41_fp_binop_rm_avx_s<bits<8> opcss, bits<8> opcsd,
5195 // Intrinsic operation, reg.
5196 def SSr_AVX : SS4AIi8<opcss, MRMSrcReg,
5197 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5198 !strconcat(OpcodeStr,
5199 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5202 // Intrinsic operation, mem.
5203 def SSm_AVX : SS4AIi8<opcss, MRMSrcMem,
5204 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
5205 !strconcat(OpcodeStr,
5206 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5209 // Intrinsic operation, reg.
5210 def SDr_AVX : SS4AIi8<opcsd, MRMSrcReg,
5211 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
5212 !strconcat(OpcodeStr,
5213 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5216 // Intrinsic operation, mem.
5217 def SDm_AVX : SS4AIi8<opcsd, MRMSrcMem,
5218 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
5219 !strconcat(OpcodeStr,
5220 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5224 // FP round - roundss, roundps, roundsd, roundpd
5225 let Predicates = [HasAVX] in {
5227 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
5228 memopv4f32, memopv2f64,
5229 int_x86_sse41_round_ps,
5230 int_x86_sse41_round_pd>, VEX;
5231 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
5232 memopv8f32, memopv4f64,
5233 int_x86_avx_round_ps_256,
5234 int_x86_avx_round_pd_256>, VEX;
5235 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
5236 int_x86_sse41_round_ss,
5237 int_x86_sse41_round_sd, 0>, VEX_4V;
5239 // Instructions for the assembler
5240 defm VROUND : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR128, f128mem, "vround">,
5242 defm VROUNDY : sse41_fp_unop_rm_avx_p<0x08, 0x09, VR256, f256mem, "vround">,
5244 defm VROUND : sse41_fp_binop_rm_avx_s<0x0A, 0x0B, "vround">, VEX_4V;
5247 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
5248 memopv4f32, memopv2f64,
5249 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
5250 let Constraints = "$src1 = $dst" in
5251 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
5252 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
5254 //===----------------------------------------------------------------------===//
5255 // SSE4.1 - Packed Bit Test
5256 //===----------------------------------------------------------------------===//
5258 // ptest instruction we'll lower to this in X86ISelLowering primarily from
5259 // the intel intrinsic that corresponds to this.
5260 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5261 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5262 "vptest\t{$src2, $src1|$src1, $src2}",
5263 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5265 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5266 "vptest\t{$src2, $src1|$src1, $src2}",
5267 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5270 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
5271 "vptest\t{$src2, $src1|$src1, $src2}",
5272 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
5274 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
5275 "vptest\t{$src2, $src1|$src1, $src2}",
5276 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
5280 let Defs = [EFLAGS] in {
5281 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
5282 "ptest \t{$src2, $src1|$src1, $src2}",
5283 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
5285 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
5286 "ptest \t{$src2, $src1|$src1, $src2}",
5287 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
5291 // The bit test instructions below are AVX only
5292 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
5293 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
5294 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
5295 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5296 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
5297 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
5298 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
5299 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
5303 let Defs = [EFLAGS], Predicates = [HasAVX] in {
5304 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
5305 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
5306 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
5307 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
5310 //===----------------------------------------------------------------------===//
5311 // SSE4.1 - Misc Instructions
5312 //===----------------------------------------------------------------------===//
5314 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
5315 "popcnt{w}\t{$src, $dst|$dst, $src}",
5316 [(set GR16:$dst, (ctpop GR16:$src))]>, OpSize, XS;
5317 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
5318 "popcnt{w}\t{$src, $dst|$dst, $src}",
5319 [(set GR16:$dst, (ctpop (loadi16 addr:$src)))]>, OpSize, XS;
5321 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
5322 "popcnt{l}\t{$src, $dst|$dst, $src}",
5323 [(set GR32:$dst, (ctpop GR32:$src))]>, XS;
5324 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
5325 "popcnt{l}\t{$src, $dst|$dst, $src}",
5326 [(set GR32:$dst, (ctpop (loadi32 addr:$src)))]>, XS;
5328 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
5329 "popcnt{q}\t{$src, $dst|$dst, $src}",
5330 [(set GR64:$dst, (ctpop GR64:$src))]>, XS;
5331 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
5332 "popcnt{q}\t{$src, $dst|$dst, $src}",
5333 [(set GR64:$dst, (ctpop (loadi64 addr:$src)))]>, XS;
5337 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
5338 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
5339 Intrinsic IntId128> {
5340 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5342 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5343 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
5344 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5346 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5349 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
5352 let Predicates = [HasAVX] in
5353 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
5354 int_x86_sse41_phminposuw>, VEX;
5355 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
5356 int_x86_sse41_phminposuw>;
5358 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
5359 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
5360 Intrinsic IntId128, bit Is2Addr = 1> {
5361 let isCommutable = 1 in
5362 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5363 (ins VR128:$src1, VR128:$src2),
5365 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5367 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
5368 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5369 (ins VR128:$src1, i128mem:$src2),
5371 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5372 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5374 (IntId128 VR128:$src1,
5375 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5378 let Predicates = [HasAVX] in {
5379 let isCommutable = 0 in
5380 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
5382 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
5384 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
5386 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
5388 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
5390 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
5392 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
5394 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
5396 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
5398 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
5400 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
5403 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5404 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
5405 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5406 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
5409 let Constraints = "$src1 = $dst" in {
5410 let isCommutable = 0 in
5411 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
5412 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
5413 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
5414 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
5415 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
5416 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
5417 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
5418 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
5419 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
5420 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
5421 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
5424 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
5425 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
5426 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
5427 (PCMPEQQrm VR128:$src1, addr:$src2)>;
5429 /// SS48I_binop_rm - Simple SSE41 binary operator.
5430 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
5431 ValueType OpVT, bit Is2Addr = 1> {
5432 let isCommutable = 1 in
5433 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5434 (ins VR128:$src1, VR128:$src2),
5436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5438 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
5440 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5441 (ins VR128:$src1, i128mem:$src2),
5443 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5444 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5445 [(set VR128:$dst, (OpNode VR128:$src1,
5446 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
5450 let Predicates = [HasAVX] in
5451 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
5452 let Constraints = "$src1 = $dst" in
5453 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
5455 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
5456 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
5457 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
5458 X86MemOperand x86memop, bit Is2Addr = 1> {
5459 let isCommutable = 1 in
5460 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
5461 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
5463 !strconcat(OpcodeStr,
5464 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5465 !strconcat(OpcodeStr,
5466 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5467 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
5469 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
5470 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
5472 !strconcat(OpcodeStr,
5473 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5474 !strconcat(OpcodeStr,
5475 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5478 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
5482 let Predicates = [HasAVX] in {
5483 let isCommutable = 0 in {
5484 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
5485 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5486 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
5487 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5488 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
5489 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5490 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
5491 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
5492 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
5493 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5494 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
5495 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5497 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
5498 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5499 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
5500 VR128, memopv16i8, i128mem, 0>, VEX_4V;
5501 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
5502 VR256, memopv32i8, i256mem, 0>, VEX_4V;
5505 let Constraints = "$src1 = $dst" in {
5506 let isCommutable = 0 in {
5507 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
5508 VR128, memopv16i8, i128mem>;
5509 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
5510 VR128, memopv16i8, i128mem>;
5511 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
5512 VR128, memopv16i8, i128mem>;
5513 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
5514 VR128, memopv16i8, i128mem>;
5516 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
5517 VR128, memopv16i8, i128mem>;
5518 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
5519 VR128, memopv16i8, i128mem>;
5522 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
5523 let Predicates = [HasAVX] in {
5524 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
5525 RegisterClass RC, X86MemOperand x86memop,
5526 PatFrag mem_frag, Intrinsic IntId> {
5527 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
5528 (ins RC:$src1, RC:$src2, RC:$src3),
5529 !strconcat(OpcodeStr,
5530 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5531 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
5532 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5534 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
5535 (ins RC:$src1, x86memop:$src2, RC:$src3),
5536 !strconcat(OpcodeStr,
5537 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5539 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
5541 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
5545 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
5546 memopv16i8, int_x86_sse41_blendvpd>;
5547 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
5548 memopv16i8, int_x86_sse41_blendvps>;
5549 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
5550 memopv16i8, int_x86_sse41_pblendvb>;
5551 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
5552 memopv32i8, int_x86_avx_blendv_pd_256>;
5553 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
5554 memopv32i8, int_x86_avx_blendv_ps_256>;
5556 /// SS41I_ternary_int - SSE 4.1 ternary operator
5557 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
5558 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5559 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
5560 (ins VR128:$src1, VR128:$src2),
5561 !strconcat(OpcodeStr,
5562 "\t{$src2, $dst|$dst, $src2}"),
5563 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
5566 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
5567 (ins VR128:$src1, i128mem:$src2),
5568 !strconcat(OpcodeStr,
5569 "\t{$src2, $dst|$dst, $src2}"),
5572 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
5576 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
5577 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
5578 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
5580 def : Pat<(X86pblendv VR128:$src1, VR128:$src2, XMM0),
5581 (PBLENDVBrr0 VR128:$src1, VR128:$src2)>;
5583 let Predicates = [HasAVX] in
5584 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5585 "vmovntdqa\t{$src, $dst|$dst, $src}",
5586 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5588 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5589 "movntdqa\t{$src, $dst|$dst, $src}",
5590 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
5593 //===----------------------------------------------------------------------===//
5594 // SSE4.2 - Compare Instructions
5595 //===----------------------------------------------------------------------===//
5597 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
5598 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
5599 Intrinsic IntId128, bit Is2Addr = 1> {
5600 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
5601 (ins VR128:$src1, VR128:$src2),
5603 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5604 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5605 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5607 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
5608 (ins VR128:$src1, i128mem:$src2),
5610 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5611 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5613 (IntId128 VR128:$src1,
5614 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5617 let Predicates = [HasAVX] in {
5618 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
5621 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5622 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
5623 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5624 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
5627 let Constraints = "$src1 = $dst" in
5628 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
5630 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
5631 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
5632 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
5633 (PCMPGTQrm VR128:$src1, addr:$src2)>;
5635 //===----------------------------------------------------------------------===//
5636 // SSE4.2 - String/text Processing Instructions
5637 //===----------------------------------------------------------------------===//
5639 // Packed Compare Implicit Length Strings, Return Mask
5640 multiclass pseudo_pcmpistrm<string asm> {
5641 def REG : PseudoI<(outs VR128:$dst),
5642 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5643 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
5645 def MEM : PseudoI<(outs VR128:$dst),
5646 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5647 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
5648 VR128:$src1, (load addr:$src2), imm:$src3))]>;
5651 let Defs = [EFLAGS], usesCustomInserter = 1 in {
5652 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
5653 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
5656 let Defs = [XMM0, EFLAGS], Predicates = [HasAVX] in {
5657 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5658 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5659 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5660 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5661 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5662 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
5665 let Defs = [XMM0, EFLAGS] in {
5666 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
5667 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5668 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5669 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
5670 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5671 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
5674 // Packed Compare Explicit Length Strings, Return Mask
5675 multiclass pseudo_pcmpestrm<string asm> {
5676 def REG : PseudoI<(outs VR128:$dst),
5677 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5678 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5679 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
5680 def MEM : PseudoI<(outs VR128:$dst),
5681 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5682 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
5683 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
5686 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
5687 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
5688 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
5691 let Predicates = [HasAVX],
5692 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5693 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5694 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5695 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5696 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5697 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5698 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
5701 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX] in {
5702 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
5703 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5704 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5705 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
5706 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5707 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
5710 // Packed Compare Implicit Length Strings, Return Index
5711 let Defs = [ECX, EFLAGS] in {
5712 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
5713 def rr : SS42AI<0x63, MRMSrcReg, (outs),
5714 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5715 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5716 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
5717 (implicit EFLAGS)]>, OpSize;
5718 def rm : SS42AI<0x63, MRMSrcMem, (outs),
5719 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5720 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
5721 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
5722 (implicit EFLAGS)]>, OpSize;
5726 let Predicates = [HasAVX] in {
5727 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
5729 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
5731 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
5733 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
5735 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
5737 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
5741 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
5742 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
5743 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
5744 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
5745 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
5746 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
5748 // Packed Compare Explicit Length Strings, Return Index
5749 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
5750 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
5751 def rr : SS42AI<0x61, MRMSrcReg, (outs),
5752 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
5753 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5754 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
5755 (implicit EFLAGS)]>, OpSize;
5756 def rm : SS42AI<0x61, MRMSrcMem, (outs),
5757 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
5758 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
5760 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
5761 (implicit EFLAGS)]>, OpSize;
5765 let Predicates = [HasAVX] in {
5766 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
5768 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
5770 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
5772 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
5774 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
5776 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
5780 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
5781 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
5782 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
5783 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
5784 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
5785 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
5787 //===----------------------------------------------------------------------===//
5788 // SSE4.2 - CRC Instructions
5789 //===----------------------------------------------------------------------===//
5791 // No CRC instructions have AVX equivalents
5793 // crc intrinsic instruction
5794 // This set of instructions are only rm, the only difference is the size
5796 let Constraints = "$src1 = $dst" in {
5797 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
5798 (ins GR32:$src1, i8mem:$src2),
5799 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5801 (int_x86_sse42_crc32_32_8 GR32:$src1,
5802 (load addr:$src2)))]>;
5803 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
5804 (ins GR32:$src1, GR8:$src2),
5805 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5807 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
5808 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5809 (ins GR32:$src1, i16mem:$src2),
5810 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5812 (int_x86_sse42_crc32_32_16 GR32:$src1,
5813 (load addr:$src2)))]>,
5815 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5816 (ins GR32:$src1, GR16:$src2),
5817 "crc32{w} \t{$src2, $src1|$src1, $src2}",
5819 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
5821 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
5822 (ins GR32:$src1, i32mem:$src2),
5823 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5825 (int_x86_sse42_crc32_32_32 GR32:$src1,
5826 (load addr:$src2)))]>;
5827 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
5828 (ins GR32:$src1, GR32:$src2),
5829 "crc32{l} \t{$src2, $src1|$src1, $src2}",
5831 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
5832 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
5833 (ins GR64:$src1, i8mem:$src2),
5834 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5836 (int_x86_sse42_crc32_64_8 GR64:$src1,
5837 (load addr:$src2)))]>,
5839 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
5840 (ins GR64:$src1, GR8:$src2),
5841 "crc32{b} \t{$src2, $src1|$src1, $src2}",
5843 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
5845 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
5846 (ins GR64:$src1, i64mem:$src2),
5847 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5849 (int_x86_sse42_crc32_64_64 GR64:$src1,
5850 (load addr:$src2)))]>,
5852 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
5853 (ins GR64:$src1, GR64:$src2),
5854 "crc32{q} \t{$src2, $src1|$src1, $src2}",
5856 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
5860 //===----------------------------------------------------------------------===//
5861 // AES-NI Instructions
5862 //===----------------------------------------------------------------------===//
5864 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
5865 Intrinsic IntId128, bit Is2Addr = 1> {
5866 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
5867 (ins VR128:$src1, VR128:$src2),
5869 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5871 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5873 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
5874 (ins VR128:$src1, i128mem:$src2),
5876 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5877 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5879 (IntId128 VR128:$src1,
5880 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
5883 // Perform One Round of an AES Encryption/Decryption Flow
5884 let Predicates = [HasAVX, HasAES] in {
5885 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
5886 int_x86_aesni_aesenc, 0>, VEX_4V;
5887 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
5888 int_x86_aesni_aesenclast, 0>, VEX_4V;
5889 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
5890 int_x86_aesni_aesdec, 0>, VEX_4V;
5891 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
5892 int_x86_aesni_aesdeclast, 0>, VEX_4V;
5895 let Constraints = "$src1 = $dst" in {
5896 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
5897 int_x86_aesni_aesenc>;
5898 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
5899 int_x86_aesni_aesenclast>;
5900 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
5901 int_x86_aesni_aesdec>;
5902 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
5903 int_x86_aesni_aesdeclast>;
5906 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
5907 (AESENCrr VR128:$src1, VR128:$src2)>;
5908 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
5909 (AESENCrm VR128:$src1, addr:$src2)>;
5910 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
5911 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
5912 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
5913 (AESENCLASTrm VR128:$src1, addr:$src2)>;
5914 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
5915 (AESDECrr VR128:$src1, VR128:$src2)>;
5916 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
5917 (AESDECrm VR128:$src1, addr:$src2)>;
5918 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
5919 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
5920 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
5921 (AESDECLASTrm VR128:$src1, addr:$src2)>;
5923 // Perform the AES InvMixColumn Transformation
5924 let Predicates = [HasAVX, HasAES] in {
5925 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5927 "vaesimc\t{$src1, $dst|$dst, $src1}",
5929 (int_x86_aesni_aesimc VR128:$src1))]>,
5931 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5932 (ins i128mem:$src1),
5933 "vaesimc\t{$src1, $dst|$dst, $src1}",
5935 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5938 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
5940 "aesimc\t{$src1, $dst|$dst, $src1}",
5942 (int_x86_aesni_aesimc VR128:$src1))]>,
5944 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
5945 (ins i128mem:$src1),
5946 "aesimc\t{$src1, $dst|$dst, $src1}",
5948 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
5951 // AES Round Key Generation Assist
5952 let Predicates = [HasAVX, HasAES] in {
5953 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5954 (ins VR128:$src1, i8imm:$src2),
5955 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5957 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5959 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5960 (ins i128mem:$src1, i8imm:$src2),
5961 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5963 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5967 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
5968 (ins VR128:$src1, i8imm:$src2),
5969 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5971 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
5973 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
5974 (ins i128mem:$src1, i8imm:$src2),
5975 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
5977 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
5981 //===----------------------------------------------------------------------===//
5982 // CLMUL Instructions
5983 //===----------------------------------------------------------------------===//
5985 // Carry-less Multiplication instructions
5986 let Constraints = "$src1 = $dst" in {
5987 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
5988 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5989 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5992 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
5993 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5994 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
5998 // AVX carry-less Multiplication instructions
5999 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
6000 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6001 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6004 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
6005 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6006 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6010 multiclass pclmul_alias<string asm, int immop> {
6011 def : InstAlias<!strconcat("pclmul", asm,
6012 "dq {$src, $dst|$dst, $src}"),
6013 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
6015 def : InstAlias<!strconcat("pclmul", asm,
6016 "dq {$src, $dst|$dst, $src}"),
6017 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
6019 def : InstAlias<!strconcat("vpclmul", asm,
6020 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6021 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
6023 def : InstAlias<!strconcat("vpclmul", asm,
6024 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
6025 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
6027 defm : pclmul_alias<"hqhq", 0x11>;
6028 defm : pclmul_alias<"hqlq", 0x01>;
6029 defm : pclmul_alias<"lqhq", 0x10>;
6030 defm : pclmul_alias<"lqlq", 0x00>;
6032 //===----------------------------------------------------------------------===//
6034 //===----------------------------------------------------------------------===//
6036 //===----------------------------------------------------------------------===//
6037 // VBROADCAST - Load from memory and broadcast to all elements of the
6038 // destination operand
6040 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
6041 X86MemOperand x86memop, Intrinsic Int> :
6042 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
6043 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6044 [(set RC:$dst, (Int addr:$src))]>, VEX;
6046 def VBROADCASTSS : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
6047 int_x86_avx_vbroadcastss>;
6048 def VBROADCASTSSY : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
6049 int_x86_avx_vbroadcastss_256>;
6050 def VBROADCASTSD : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
6051 int_x86_avx_vbroadcast_sd_256>;
6052 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
6053 int_x86_avx_vbroadcastf128_pd_256>;
6055 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
6056 (VBROADCASTF128 addr:$src)>;
6058 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
6059 (VBROADCASTSSY addr:$src)>;
6060 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
6061 (VBROADCASTSD addr:$src)>;
6062 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
6063 (VBROADCASTSSY addr:$src)>;
6064 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
6065 (VBROADCASTSD addr:$src)>;
6067 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
6068 (VBROADCASTSS addr:$src)>;
6069 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
6070 (VBROADCASTSS addr:$src)>;
6072 //===----------------------------------------------------------------------===//
6073 // VINSERTF128 - Insert packed floating-point values
6075 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
6076 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
6077 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6079 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
6080 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
6081 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6084 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
6085 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6086 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
6087 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6088 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
6089 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
6091 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
6093 (VINSERTF128rr VR256:$src1, VR128:$src2,
6094 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6095 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
6097 (VINSERTF128rr VR256:$src1, VR128:$src2,
6098 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6099 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
6101 (VINSERTF128rr VR256:$src1, VR128:$src2,
6102 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6103 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
6105 (VINSERTF128rr VR256:$src1, VR128:$src2,
6106 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6107 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
6109 (VINSERTF128rr VR256:$src1, VR128:$src2,
6110 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6111 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
6113 (VINSERTF128rr VR256:$src1, VR128:$src2,
6114 (INSERT_get_vinsertf128_imm VR256:$ins))>;
6116 //===----------------------------------------------------------------------===//
6117 // VEXTRACTF128 - Extract packed floating-point values
6119 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
6120 (ins VR256:$src1, i8imm:$src2),
6121 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6123 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
6124 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
6125 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
6128 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
6129 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6130 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
6131 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6132 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
6133 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
6135 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6136 (v4f32 (VEXTRACTF128rr
6137 (v8f32 VR256:$src1),
6138 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6139 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6140 (v2f64 (VEXTRACTF128rr
6141 (v4f64 VR256:$src1),
6142 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6143 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6144 (v4i32 (VEXTRACTF128rr
6145 (v8i32 VR256:$src1),
6146 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6147 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6148 (v2i64 (VEXTRACTF128rr
6149 (v4i64 VR256:$src1),
6150 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6151 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6152 (v8i16 (VEXTRACTF128rr
6153 (v16i16 VR256:$src1),
6154 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6155 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
6156 (v16i8 (VEXTRACTF128rr
6157 (v32i8 VR256:$src1),
6158 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
6160 //===----------------------------------------------------------------------===//
6161 // VMASKMOV - Conditional SIMD Packed Loads and Stores
6163 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
6164 Intrinsic IntLd, Intrinsic IntLd256,
6165 Intrinsic IntSt, Intrinsic IntSt256,
6166 PatFrag pf128, PatFrag pf256> {
6167 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
6168 (ins VR128:$src1, f128mem:$src2),
6169 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6170 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
6172 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
6173 (ins VR256:$src1, f256mem:$src2),
6174 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6175 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
6177 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
6178 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
6179 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6180 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
6181 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
6182 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
6183 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6184 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
6187 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
6188 int_x86_avx_maskload_ps,
6189 int_x86_avx_maskload_ps_256,
6190 int_x86_avx_maskstore_ps,
6191 int_x86_avx_maskstore_ps_256,
6192 memopv4f32, memopv8f32>;
6193 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
6194 int_x86_avx_maskload_pd,
6195 int_x86_avx_maskload_pd_256,
6196 int_x86_avx_maskstore_pd,
6197 int_x86_avx_maskstore_pd_256,
6198 memopv2f64, memopv4f64>;
6200 //===----------------------------------------------------------------------===//
6201 // VPERMIL - Permute Single and Double Floating-Point Values
6203 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
6204 RegisterClass RC, X86MemOperand x86memop_f,
6205 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
6206 Intrinsic IntVar, Intrinsic IntImm> {
6207 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
6208 (ins RC:$src1, RC:$src2),
6209 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6210 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
6211 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
6212 (ins RC:$src1, x86memop_i:$src2),
6213 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6214 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
6216 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
6217 (ins RC:$src1, i8imm:$src2),
6218 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6219 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
6220 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
6221 (ins x86memop_f:$src1, i8imm:$src2),
6222 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6223 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
6226 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
6227 memopv4f32, memopv4i32,
6228 int_x86_avx_vpermilvar_ps,
6229 int_x86_avx_vpermil_ps>;
6230 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
6231 memopv8f32, memopv8i32,
6232 int_x86_avx_vpermilvar_ps_256,
6233 int_x86_avx_vpermil_ps_256>;
6234 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
6235 memopv2f64, memopv2i64,
6236 int_x86_avx_vpermilvar_pd,
6237 int_x86_avx_vpermil_pd>;
6238 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
6239 memopv4f64, memopv4i64,
6240 int_x86_avx_vpermilvar_pd_256,
6241 int_x86_avx_vpermil_pd_256>;
6243 def : Pat<(v8f32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6244 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6245 def : Pat<(v4f64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6246 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6247 def : Pat<(v8i32 (X86VPermilpsy VR256:$src1, (i8 imm:$imm))),
6248 (VPERMILPSYri VR256:$src1, imm:$imm)>;
6249 def : Pat<(v4i64 (X86VPermilpdy VR256:$src1, (i8 imm:$imm))),
6250 (VPERMILPDYri VR256:$src1, imm:$imm)>;
6252 //===----------------------------------------------------------------------===//
6253 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
6255 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
6256 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
6257 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6259 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
6260 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
6261 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
6264 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
6265 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6266 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
6267 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6268 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
6269 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
6271 def : Pat<(int_x86_avx_vperm2f128_ps_256
6272 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
6273 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6274 def : Pat<(int_x86_avx_vperm2f128_pd_256
6275 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
6276 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6277 def : Pat<(int_x86_avx_vperm2f128_si_256
6278 VR256:$src1, (memopv8i32 addr:$src2), imm:$src3),
6279 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
6281 def : Pat<(v8f32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6282 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6283 def : Pat<(v8i32 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6284 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6285 def : Pat<(v4i64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6286 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6287 def : Pat<(v4f64 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6288 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6289 def : Pat<(v32i8 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6290 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6291 def : Pat<(v16i16 (X86VPerm2f128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
6292 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
6294 //===----------------------------------------------------------------------===//
6295 // VZERO - Zero YMM registers
6297 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
6298 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
6299 // Zero All YMM registers
6300 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
6301 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
6303 // Zero Upper bits of YMM registers
6304 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
6305 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
6308 //===----------------------------------------------------------------------===//
6309 // SSE Shuffle pattern fragments
6310 //===----------------------------------------------------------------------===//
6312 // This is part of a "work in progress" refactoring. The idea is that all
6313 // vector shuffles are going to be translated into target specific nodes and
6314 // directly matched by the patterns below (which can be changed along the way)
6315 // The AVX version of some but not all of them are described here, and more
6316 // should come in a near future.
6318 // Shuffle with MOVLHPD
6319 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
6320 (scalar_to_vector (loadf64 addr:$src2)))),
6321 (MOVHPDrm VR128:$src1, addr:$src2)>;
6323 // FIXME: Instead of X86Unpcklpd, there should be a X86Movlhpd here, the problem
6324 // is during lowering, where it's not possible to recognize the load fold cause
6325 // it has two uses through a bitcast. One use disappears at isel time and the
6326 // fold opportunity reappears.
6327 def : Pat<(v2f64 (X86Unpcklpd VR128:$src1,
6328 (scalar_to_vector (loadf64 addr:$src2)))),
6329 (MOVHPDrm VR128:$src1, addr:$src2)>;
6331 // Shuffle with MOVLPS
6332 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
6333 (MOVLPSrm VR128:$src1, addr:$src2)>;
6334 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
6335 (MOVLPSrm VR128:$src1, addr:$src2)>;
6336 def : Pat<(X86Movlps VR128:$src1,
6337 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
6338 (MOVLPSrm VR128:$src1, addr:$src2)>;
6340 // Shuffle with MOVLPD
6341 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6342 (MOVLPDrm VR128:$src1, addr:$src2)>;
6343 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
6344 (MOVLPDrm VR128:$src1, addr:$src2)>;
6345 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
6346 (scalar_to_vector (loadf64 addr:$src2)))),
6347 (MOVLPDrm VR128:$src1, addr:$src2)>;
6349 // Extra patterns to match stores with MOVHPS/PD and MOVLPS/PD
6350 def : Pat<(store (f64 (vector_extract
6351 (v2f64 (X86Unpckhps VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6352 (MOVHPSmr addr:$dst, VR128:$src)>;
6353 def : Pat<(store (f64 (vector_extract
6354 (v2f64 (X86Unpckhpd VR128:$src, (undef))), (iPTR 0))),addr:$dst),
6355 (MOVHPDmr addr:$dst, VR128:$src)>;
6357 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),addr:$src1),
6358 (MOVLPSmr addr:$src1, VR128:$src2)>;
6359 def : Pat<(store (v4i32 (X86Movlps
6360 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
6361 (MOVLPSmr addr:$src1, VR128:$src2)>;
6363 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6364 (MOVLPDmr addr:$src1, VR128:$src2)>;
6365 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),addr:$src1),
6366 (MOVLPDmr addr:$src1, VR128:$src2)>;