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 bit rr_hasSideEffects = 0> {
85 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
86 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
88 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
89 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
93 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
94 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
98 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
99 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
100 string asm, string SSEVer, string FPSizeStr,
101 X86MemOperand x86memop, PatFrag mem_frag,
102 Domain d, bit Is2Addr = 1> {
103 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
105 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
106 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
107 [(set RC:$dst, (!cast<Intrinsic>(
108 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
109 RC:$src1, RC:$src2))], d>;
110 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
112 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
113 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
114 [(set RC:$dst, (!cast<Intrinsic>(
115 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
116 RC:$src1, (mem_frag addr:$src2)))], d>;
119 //===----------------------------------------------------------------------===//
120 // Non-instruction patterns
121 //===----------------------------------------------------------------------===//
123 // A vector extract of the first f32/f64 position is a subregister copy
124 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
125 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
127 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
129 // A 128-bit subvector extract from the first 256-bit vector position
130 // is a subregister copy that needs no instruction.
131 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
132 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
133 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
134 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
136 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
137 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
138 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
139 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
141 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
142 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
143 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
144 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
146 // A 128-bit subvector insert to the first 256-bit vector position
147 // is a subregister copy that needs no instruction.
148 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
149 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
150 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
151 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
152 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
153 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
154 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
155 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
156 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
157 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
159 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
161 // Implicitly promote a 32-bit scalar to a vector.
162 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
163 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
164 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
165 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
166 // Implicitly promote a 64-bit scalar to a vector.
167 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
168 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
170 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
172 // Bitcasts between 128-bit vector types. Return the original type since
173 // no instruction is needed for the conversion
174 let Predicates = [HasXMMInt] in {
175 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
207 // Bitcasts between 256-bit vector types. Return the original type since
208 // no instruction is needed for the conversion
209 let Predicates = [HasAVX] in {
210 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
242 // Alias instructions that map fld0 to pxor for sse.
243 // This is expanded by ExpandPostRAPseudos.
244 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
246 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
247 [(set FR32:$dst, fp32imm0)]>, Requires<[HasXMM]>;
248 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
249 [(set FR64:$dst, fpimm0)]>, Requires<[HasXMMInt]>;
252 //===----------------------------------------------------------------------===//
253 // AVX & SSE - Zero/One Vectors
254 //===----------------------------------------------------------------------===//
256 // Alias instruction that maps zero vector to pxor / xorp* for sse.
257 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
258 // swizzled by ExecutionDepsFix to pxor.
259 // We set canFoldAsLoad because this can be converted to a constant-pool
260 // load of an all-zeros value if folding it would be beneficial.
261 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
262 isPseudo = 1, neverHasSideEffects = 1 in {
263 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
266 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
267 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
268 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
269 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
270 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
271 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
274 // The same as done above but for AVX. The 256-bit ISA does not support PI,
275 // and doesn't need it because on sandy bridge the register is set to zero
276 // at the rename stage without using any execution unit, so SET0PSY
277 // and SET0PDY can be used for vector int instructions without penalty
278 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
279 // JIT implementatioan, it does not expand the instructions below like
280 // X86MCInstLower does.
281 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
282 isCodeGenOnly = 1, Predicates = [HasAVX] in {
283 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
284 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
285 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 // AVX has no support for 256-bit integer instructions, but since the 128-bit
291 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
292 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
293 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
294 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
296 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
297 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
298 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
300 // We set canFoldAsLoad because this can be converted to a constant-pool
301 // load of an all-ones value if folding it would be beneficial.
302 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
303 // JIT implementation, it does not expand the instructions below like
304 // X86MCInstLower does.
305 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
306 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
307 let Predicates = [HasAVX] in
308 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
309 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
310 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
311 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
312 let Predicates = [HasAVX2] in
313 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
314 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
318 //===----------------------------------------------------------------------===//
319 // SSE 1 & 2 - Move FP Scalar Instructions
321 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
322 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
323 // is used instead. Register-to-register movss/movsd is not modeled as an
324 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
325 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
326 //===----------------------------------------------------------------------===//
328 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
329 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
330 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
332 // Loading from memory automatically zeroing upper bits.
333 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
334 PatFrag mem_pat, string OpcodeStr> :
335 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
337 [(set RC:$dst, (mem_pat addr:$src))]>;
340 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
341 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
343 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
344 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
347 // For the disassembler
348 let isCodeGenOnly = 1 in {
349 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
350 (ins VR128:$src1, FR32:$src2),
351 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
353 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
354 (ins VR128:$src1, FR64:$src2),
355 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
359 let canFoldAsLoad = 1, isReMaterializable = 1 in {
360 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
362 let AddedComplexity = 20 in
363 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
367 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
368 "movss\t{$src, $dst|$dst, $src}",
369 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
370 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
371 "movsd\t{$src, $dst|$dst, $src}",
372 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
375 let Constraints = "$src1 = $dst" in {
376 def MOVSSrr : sse12_move_rr<FR32, v4f32,
377 "movss\t{$src2, $dst|$dst, $src2}">, XS;
378 def MOVSDrr : sse12_move_rr<FR64, v2f64,
379 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
381 // For the disassembler
382 let isCodeGenOnly = 1 in {
383 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
384 (ins VR128:$src1, FR32:$src2),
385 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
386 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
387 (ins VR128:$src1, FR64:$src2),
388 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
392 let canFoldAsLoad = 1, isReMaterializable = 1 in {
393 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
395 let AddedComplexity = 20 in
396 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
399 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
400 "movss\t{$src, $dst|$dst, $src}",
401 [(store FR32:$src, addr:$dst)]>;
402 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
403 "movsd\t{$src, $dst|$dst, $src}",
404 [(store FR64:$src, addr:$dst)]>;
407 let Predicates = [HasAVX] in {
408 let AddedComplexity = 15 in {
409 // Extract the low 32-bit value from one vector and insert it into another.
410 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
411 (VMOVSSrr (v4f32 VR128:$src1),
412 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
413 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
414 (VMOVSSrr (v4i32 VR128:$src1),
415 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
417 // Extract the low 64-bit value from one vector and insert it into another.
418 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
419 (VMOVSDrr (v2f64 VR128:$src1),
420 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
421 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
422 (VMOVSDrr (v2i64 VR128:$src1),
423 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
425 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
426 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
427 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
428 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
429 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
431 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
432 // MOVS{S,D} to the lower bits.
433 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
434 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
435 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
436 (VMOVSSrr (v4f32 (V_SET0)),
437 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
438 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
439 (VMOVSSrr (v4i32 (V_SET0)),
440 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
441 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
442 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
444 // Move low f32 and clear high bits.
445 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
446 (SUBREG_TO_REG (i32 0),
447 (VMOVSSrr (v4f32 (V_SET0)),
448 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
449 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
450 (SUBREG_TO_REG (i32 0),
451 (VMOVSSrr (v4i32 (V_SET0)),
452 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
455 let AddedComplexity = 20 in {
456 // MOVSSrm zeros the high parts of the register; represent this
457 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
458 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
459 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
460 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
461 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
462 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
463 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
465 // MOVSDrm zeros the high parts of the register; represent this
466 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
467 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
468 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
469 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
470 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
471 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
472 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
473 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
474 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
475 def : Pat<(v2f64 (X86vzload addr:$src)),
476 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
478 // Represent the same patterns above but in the form they appear for
480 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
481 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
482 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
483 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
484 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
485 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
486 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
487 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
488 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
490 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
491 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
492 (SUBREG_TO_REG (i32 0),
493 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
495 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
496 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
497 (SUBREG_TO_REG (i64 0),
498 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
500 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
501 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
502 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
504 // Move low f64 and clear high bits.
505 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
506 (SUBREG_TO_REG (i32 0),
507 (VMOVSDrr (v2f64 (V_SET0)),
508 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
510 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
511 (SUBREG_TO_REG (i32 0),
512 (VMOVSDrr (v2i64 (V_SET0)),
513 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
515 // Extract and store.
516 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
519 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
520 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
523 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
525 // Shuffle with VMOVSS
526 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
527 (VMOVSSrr VR128:$src1, FR32:$src2)>;
528 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
529 (VMOVSSrr (v4i32 VR128:$src1),
530 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
531 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
532 (VMOVSSrr (v4f32 VR128:$src1),
533 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
536 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
537 (SUBREG_TO_REG (i32 0),
538 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
539 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
540 def : Pat<(v8f32 (X86Movsd VR256:$src1, VR256:$src2)),
541 (SUBREG_TO_REG (i32 0),
542 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
543 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
545 // Shuffle with VMOVSD
546 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
547 (VMOVSDrr VR128:$src1, FR64:$src2)>;
548 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
549 (VMOVSDrr (v2i64 VR128:$src1),
550 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
551 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
552 (VMOVSDrr (v2f64 VR128:$src1),
553 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
554 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
555 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
557 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
558 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
562 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
563 (SUBREG_TO_REG (i32 0),
564 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
565 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
566 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
567 (SUBREG_TO_REG (i32 0),
568 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
569 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
572 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
573 // is during lowering, where it's not possible to recognize the fold cause
574 // it has two uses through a bitcast. One use disappears at isel time and the
575 // fold opportunity reappears.
576 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
577 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
579 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
580 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
582 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
583 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
585 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
586 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
590 let Predicates = [HasSSE1] in {
591 let AddedComplexity = 15 in {
592 // Extract the low 32-bit value from one vector and insert it into another.
593 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
594 (MOVSSrr (v4f32 VR128:$src1),
595 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
596 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
597 (MOVSSrr (v4i32 VR128:$src1),
598 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
600 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
601 // MOVSS to the lower bits.
602 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
603 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
604 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
605 (MOVSSrr (v4f32 (V_SET0)),
606 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
607 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
608 (MOVSSrr (v4i32 (V_SET0)),
609 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
612 let AddedComplexity = 20 in {
613 // MOVSSrm zeros the high parts of the register; represent this
614 // with SUBREG_TO_REG.
615 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
616 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
617 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
618 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
619 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
620 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
623 // Extract and store.
624 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
627 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
629 // Shuffle with MOVSS
630 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
631 (MOVSSrr VR128:$src1, FR32:$src2)>;
632 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
633 (MOVSSrr (v4i32 VR128:$src1),
634 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
635 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
636 (MOVSSrr (v4f32 VR128:$src1),
637 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
640 let Predicates = [HasSSE2] in {
641 let AddedComplexity = 15 in {
642 // Extract the low 64-bit value from one vector and insert it into another.
643 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
644 (MOVSDrr (v2f64 VR128:$src1),
645 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
646 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
647 (MOVSDrr (v2i64 VR128:$src1),
648 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
650 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
651 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
652 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
653 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
654 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
656 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
657 // MOVSD to the lower bits.
658 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
659 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
662 let AddedComplexity = 20 in {
663 // MOVSDrm zeros the high parts of the register; represent this
664 // with SUBREG_TO_REG.
665 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
666 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
667 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
668 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
669 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
670 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
671 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
672 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
673 def : Pat<(v2f64 (X86vzload addr:$src)),
674 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
677 // Extract and store.
678 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
681 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
683 // Shuffle with MOVSD
684 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
685 (MOVSDrr VR128:$src1, FR64:$src2)>;
686 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
687 (MOVSDrr (v2i64 VR128:$src1),
688 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
689 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
690 (MOVSDrr (v2f64 VR128:$src1),
691 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
692 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
693 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
694 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
695 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
697 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
698 // is during lowering, where it's not possible to recognize the fold cause
699 // it has two uses through a bitcast. One use disappears at isel time and the
700 // fold opportunity reappears.
701 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
702 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
703 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
704 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
705 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
706 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
707 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
708 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
711 //===----------------------------------------------------------------------===//
712 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
713 //===----------------------------------------------------------------------===//
715 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
716 X86MemOperand x86memop, PatFrag ld_frag,
717 string asm, Domain d,
718 bit IsReMaterializable = 1> {
719 let neverHasSideEffects = 1 in
720 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
721 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
722 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
723 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
724 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
725 [(set RC:$dst, (ld_frag addr:$src))], d>;
728 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
729 "movaps", SSEPackedSingle>, TB, VEX;
730 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
731 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
732 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
733 "movups", SSEPackedSingle>, TB, VEX;
734 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
735 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
737 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
738 "movaps", SSEPackedSingle>, TB, VEX;
739 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
740 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
741 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
742 "movups", SSEPackedSingle>, TB, VEX;
743 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
744 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
745 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
746 "movaps", SSEPackedSingle>, TB;
747 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
748 "movapd", SSEPackedDouble>, TB, OpSize;
749 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
750 "movups", SSEPackedSingle>, TB;
751 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
752 "movupd", SSEPackedDouble, 0>, TB, OpSize;
754 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
755 "movaps\t{$src, $dst|$dst, $src}",
756 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
757 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
758 "movapd\t{$src, $dst|$dst, $src}",
759 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
760 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
761 "movups\t{$src, $dst|$dst, $src}",
762 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
763 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
764 "movupd\t{$src, $dst|$dst, $src}",
765 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
766 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
767 "movaps\t{$src, $dst|$dst, $src}",
768 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
769 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
770 "movapd\t{$src, $dst|$dst, $src}",
771 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
772 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
773 "movups\t{$src, $dst|$dst, $src}",
774 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
775 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
776 "movupd\t{$src, $dst|$dst, $src}",
777 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
780 let isCodeGenOnly = 1 in {
781 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
783 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
784 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
786 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
787 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
789 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
790 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
792 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
793 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
795 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
796 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
798 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
799 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
801 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
802 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
804 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
807 let Predicates = [HasAVX] in {
808 def : Pat<(v8i32 (X86vzmovl
809 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
810 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
811 def : Pat<(v4i64 (X86vzmovl
812 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
813 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
814 def : Pat<(v8f32 (X86vzmovl
815 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
816 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
817 def : Pat<(v4f64 (X86vzmovl
818 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
819 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
823 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
824 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
825 (VMOVUPSYmr addr:$dst, VR256:$src)>;
827 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
828 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
829 (VMOVUPDYmr addr:$dst, VR256:$src)>;
831 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
832 "movaps\t{$src, $dst|$dst, $src}",
833 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
834 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
835 "movapd\t{$src, $dst|$dst, $src}",
836 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
837 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
838 "movups\t{$src, $dst|$dst, $src}",
839 [(store (v4f32 VR128:$src), addr:$dst)]>;
840 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
841 "movupd\t{$src, $dst|$dst, $src}",
842 [(store (v2f64 VR128:$src), addr:$dst)]>;
845 let isCodeGenOnly = 1 in {
846 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
847 "movaps\t{$src, $dst|$dst, $src}", []>;
848 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
849 "movapd\t{$src, $dst|$dst, $src}", []>;
850 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
851 "movups\t{$src, $dst|$dst, $src}", []>;
852 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
853 "movupd\t{$src, $dst|$dst, $src}", []>;
856 let Predicates = [HasAVX] in {
857 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
858 (VMOVUPSmr addr:$dst, VR128:$src)>;
859 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
860 (VMOVUPDmr addr:$dst, VR128:$src)>;
863 let Predicates = [HasSSE1] in
864 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
865 (MOVUPSmr addr:$dst, VR128:$src)>;
866 let Predicates = [HasSSE2] in
867 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
868 (MOVUPDmr addr:$dst, VR128:$src)>;
870 // Use vmovaps/vmovups for AVX integer load/store.
871 let Predicates = [HasAVX] in {
872 // 128-bit load/store
873 def : Pat<(alignedloadv4i32 addr:$src),
874 (VMOVAPSrm addr:$src)>;
875 def : Pat<(loadv4i32 addr:$src),
876 (VMOVUPSrm addr:$src)>;
877 def : Pat<(alignedloadv2i64 addr:$src),
878 (VMOVAPSrm addr:$src)>;
879 def : Pat<(loadv2i64 addr:$src),
880 (VMOVUPSrm addr:$src)>;
882 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
883 (VMOVAPSmr addr:$dst, VR128:$src)>;
884 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
885 (VMOVAPSmr addr:$dst, VR128:$src)>;
886 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
887 (VMOVAPSmr addr:$dst, VR128:$src)>;
888 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
889 (VMOVAPSmr addr:$dst, VR128:$src)>;
890 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
891 (VMOVUPSmr addr:$dst, VR128:$src)>;
892 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
893 (VMOVUPSmr addr:$dst, VR128:$src)>;
894 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
895 (VMOVUPSmr addr:$dst, VR128:$src)>;
896 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
897 (VMOVUPSmr addr:$dst, VR128:$src)>;
899 // 256-bit load/store
900 def : Pat<(alignedloadv4i64 addr:$src),
901 (VMOVAPSYrm addr:$src)>;
902 def : Pat<(loadv4i64 addr:$src),
903 (VMOVUPSYrm addr:$src)>;
904 def : Pat<(alignedloadv8i32 addr:$src),
905 (VMOVAPSYrm addr:$src)>;
906 def : Pat<(loadv8i32 addr:$src),
907 (VMOVUPSYrm addr:$src)>;
908 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
909 (VMOVAPSYmr addr:$dst, VR256:$src)>;
910 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
911 (VMOVAPSYmr addr:$dst, VR256:$src)>;
912 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
913 (VMOVAPSYmr addr:$dst, VR256:$src)>;
914 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
915 (VMOVAPSYmr addr:$dst, VR256:$src)>;
916 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
917 (VMOVUPSYmr addr:$dst, VR256:$src)>;
918 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
919 (VMOVUPSYmr addr:$dst, VR256:$src)>;
920 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
921 (VMOVUPSYmr addr:$dst, VR256:$src)>;
922 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
923 (VMOVUPSYmr addr:$dst, VR256:$src)>;
926 // Use movaps / movups for SSE integer load / store (one byte shorter).
927 // The instructions selected below are then converted to MOVDQA/MOVDQU
928 // during the SSE domain pass.
929 let Predicates = [HasSSE1] in {
930 def : Pat<(alignedloadv4i32 addr:$src),
931 (MOVAPSrm addr:$src)>;
932 def : Pat<(loadv4i32 addr:$src),
933 (MOVUPSrm addr:$src)>;
934 def : Pat<(alignedloadv2i64 addr:$src),
935 (MOVAPSrm addr:$src)>;
936 def : Pat<(loadv2i64 addr:$src),
937 (MOVUPSrm addr:$src)>;
939 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
940 (MOVAPSmr addr:$dst, VR128:$src)>;
941 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
942 (MOVAPSmr addr:$dst, VR128:$src)>;
943 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
944 (MOVAPSmr addr:$dst, VR128:$src)>;
945 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
946 (MOVAPSmr addr:$dst, VR128:$src)>;
947 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
948 (MOVUPSmr addr:$dst, VR128:$src)>;
949 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
950 (MOVUPSmr addr:$dst, VR128:$src)>;
951 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
952 (MOVUPSmr addr:$dst, VR128:$src)>;
953 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
954 (MOVUPSmr addr:$dst, VR128:$src)>;
957 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
958 // bits are disregarded. FIXME: Set encoding to pseudo!
959 let neverHasSideEffects = 1 in {
960 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
961 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
962 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
963 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
964 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
965 "movaps\t{$src, $dst|$dst, $src}", []>;
966 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
967 "movapd\t{$src, $dst|$dst, $src}", []>;
970 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
971 // bits are disregarded. FIXME: Set encoding to pseudo!
972 let canFoldAsLoad = 1, isReMaterializable = 1 in {
973 let isCodeGenOnly = 1 in {
974 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
975 "movaps\t{$src, $dst|$dst, $src}",
976 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
977 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
978 "movapd\t{$src, $dst|$dst, $src}",
979 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
981 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
982 "movaps\t{$src, $dst|$dst, $src}",
983 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
984 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
985 "movapd\t{$src, $dst|$dst, $src}",
986 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
989 //===----------------------------------------------------------------------===//
990 // SSE 1 & 2 - Move Low packed FP Instructions
991 //===----------------------------------------------------------------------===//
993 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
994 PatFrag mov_frag, string base_opc,
996 def PSrm : PI<opc, MRMSrcMem,
997 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
998 !strconcat(base_opc, "s", asm_opr),
1001 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1002 SSEPackedSingle>, TB;
1004 def PDrm : PI<opc, MRMSrcMem,
1005 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1006 !strconcat(base_opc, "d", asm_opr),
1007 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1008 (scalar_to_vector (loadf64 addr:$src2)))))],
1009 SSEPackedDouble>, TB, OpSize;
1012 let AddedComplexity = 20 in {
1013 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1014 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1016 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1017 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1018 "\t{$src2, $dst|$dst, $src2}">;
1021 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1022 "movlps\t{$src, $dst|$dst, $src}",
1023 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1024 (iPTR 0))), addr:$dst)]>, VEX;
1025 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1026 "movlpd\t{$src, $dst|$dst, $src}",
1027 [(store (f64 (vector_extract (v2f64 VR128:$src),
1028 (iPTR 0))), addr:$dst)]>, VEX;
1029 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1030 "movlps\t{$src, $dst|$dst, $src}",
1031 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1032 (iPTR 0))), addr:$dst)]>;
1033 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1034 "movlpd\t{$src, $dst|$dst, $src}",
1035 [(store (f64 (vector_extract (v2f64 VR128:$src),
1036 (iPTR 0))), addr:$dst)]>;
1038 let Predicates = [HasAVX] in {
1039 let AddedComplexity = 20 in {
1040 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1041 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1042 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1043 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1044 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1045 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1046 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1047 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1048 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1049 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1052 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1053 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1054 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1055 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1056 VR128:$src2)), addr:$src1),
1057 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1059 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1060 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1061 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1062 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1063 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1065 // Shuffle with VMOVLPS
1066 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1067 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1068 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1069 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1070 def : Pat<(X86Movlps VR128:$src1,
1071 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1072 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1074 // Shuffle with VMOVLPD
1075 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1076 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1077 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1078 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1079 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1080 (scalar_to_vector (loadf64 addr:$src2)))),
1081 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1086 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1087 def : Pat<(store (v4i32 (X86Movlps
1088 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1089 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1090 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1092 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1093 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1095 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1098 let Predicates = [HasSSE1] in {
1099 let AddedComplexity = 20 in {
1100 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1101 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1102 (MOVLPSrm VR128:$src1, addr:$src2)>;
1103 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1104 (MOVLPSrm VR128:$src1, addr:$src2)>;
1107 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1108 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1109 (iPTR 0))), addr:$src1),
1110 (MOVLPSmr addr:$src1, VR128:$src2)>;
1111 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1112 (MOVLPSmr addr:$src1, VR128:$src2)>;
1113 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1114 VR128:$src2)), addr:$src1),
1115 (MOVLPSmr addr:$src1, VR128:$src2)>;
1117 // Shuffle with MOVLPS
1118 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1119 (MOVLPSrm VR128:$src1, addr:$src2)>;
1120 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1121 (MOVLPSrm VR128:$src1, addr:$src2)>;
1122 def : Pat<(X86Movlps VR128:$src1,
1123 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1124 (MOVLPSrm VR128:$src1, addr:$src2)>;
1125 def : Pat<(X86Movlps VR128:$src1,
1126 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1127 (MOVLPSrm VR128:$src1, addr:$src2)>;
1130 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1132 (MOVLPSmr addr:$src1, VR128:$src2)>;
1133 def : Pat<(store (v4i32 (X86Movlps
1134 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1136 (MOVLPSmr addr:$src1, VR128:$src2)>;
1139 let Predicates = [HasSSE2] in {
1140 let AddedComplexity = 20 in {
1141 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1142 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1143 (MOVLPDrm VR128:$src1, addr:$src2)>;
1144 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1145 (MOVLPDrm VR128:$src1, addr:$src2)>;
1148 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1149 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1150 (MOVLPDmr addr:$src1, VR128:$src2)>;
1151 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1152 (MOVLPDmr addr:$src1, VR128:$src2)>;
1154 // Shuffle with MOVLPD
1155 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1156 (MOVLPDrm VR128:$src1, addr:$src2)>;
1157 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1158 (MOVLPDrm VR128:$src1, addr:$src2)>;
1159 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1160 (scalar_to_vector (loadf64 addr:$src2)))),
1161 (MOVLPDrm VR128:$src1, addr:$src2)>;
1164 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1166 (MOVLPDmr addr:$src1, VR128:$src2)>;
1167 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1169 (MOVLPDmr addr:$src1, VR128:$src2)>;
1172 //===----------------------------------------------------------------------===//
1173 // SSE 1 & 2 - Move Hi packed FP Instructions
1174 //===----------------------------------------------------------------------===//
1176 let AddedComplexity = 20 in {
1177 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1178 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1180 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1181 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1182 "\t{$src2, $dst|$dst, $src2}">;
1185 // v2f64 extract element 1 is always custom lowered to unpack high to low
1186 // and extract element 0 so the non-store version isn't too horrible.
1187 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1188 "movhps\t{$src, $dst|$dst, $src}",
1189 [(store (f64 (vector_extract
1190 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1191 (undef)), (iPTR 0))), addr:$dst)]>,
1193 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1194 "movhpd\t{$src, $dst|$dst, $src}",
1195 [(store (f64 (vector_extract
1196 (v2f64 (unpckh VR128:$src, (undef))),
1197 (iPTR 0))), addr:$dst)]>,
1199 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1200 "movhps\t{$src, $dst|$dst, $src}",
1201 [(store (f64 (vector_extract
1202 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1203 (undef)), (iPTR 0))), addr:$dst)]>;
1204 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1205 "movhpd\t{$src, $dst|$dst, $src}",
1206 [(store (f64 (vector_extract
1207 (v2f64 (unpckh VR128:$src, (undef))),
1208 (iPTR 0))), addr:$dst)]>;
1210 let Predicates = [HasAVX] in {
1212 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1213 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1214 def : Pat<(X86Movlhps VR128:$src1,
1215 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1216 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1217 def : Pat<(X86Movlhps VR128:$src1,
1218 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1219 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1220 def : Pat<(X86Movlhps VR128:$src1,
1221 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1222 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1224 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1225 // is during lowering, where it's not possible to recognize the load fold
1226 // cause it has two uses through a bitcast. One use disappears at isel time
1227 // and the fold opportunity reappears.
1228 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1229 (scalar_to_vector (loadf64 addr:$src2)))),
1230 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1232 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1233 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1234 (scalar_to_vector (loadf64 addr:$src2)))),
1235 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1238 def : Pat<(store (f64 (vector_extract
1239 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1240 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1241 (VMOVHPSmr addr:$dst, VR128:$src)>;
1242 def : Pat<(store (f64 (vector_extract
1243 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1244 (VMOVHPDmr addr:$dst, VR128:$src)>;
1247 let Predicates = [HasSSE1] in {
1249 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1250 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1251 def : Pat<(X86Movlhps VR128:$src1,
1252 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1253 (MOVHPSrm VR128:$src1, addr:$src2)>;
1254 def : Pat<(X86Movlhps VR128:$src1,
1255 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1256 (MOVHPSrm VR128:$src1, addr:$src2)>;
1257 def : Pat<(X86Movlhps VR128:$src1,
1258 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1259 (MOVHPSrm VR128:$src1, addr:$src2)>;
1262 def : Pat<(store (f64 (vector_extract
1263 (X86Unpckh (bc_v2f64 (v4f32 VR128:$src)),
1264 (bc_v2f64 (v4f32 VR128:$src))), (iPTR 0))), addr:$dst),
1265 (MOVHPSmr addr:$dst, VR128:$src)>;
1268 let Predicates = [HasSSE2] in {
1269 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1270 // is during lowering, where it's not possible to recognize the load fold
1271 // cause it has two uses through a bitcast. One use disappears at isel time
1272 // and the fold opportunity reappears.
1273 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1274 (scalar_to_vector (loadf64 addr:$src2)))),
1275 (MOVHPDrm VR128:$src1, addr:$src2)>;
1277 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1278 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1279 (scalar_to_vector (loadf64 addr:$src2)))),
1280 (MOVHPDrm VR128:$src1, addr:$src2)>;
1283 def : Pat<(store (f64 (vector_extract
1284 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1285 (MOVHPDmr addr:$dst, VR128:$src)>;
1288 //===----------------------------------------------------------------------===//
1289 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1290 //===----------------------------------------------------------------------===//
1292 let AddedComplexity = 20 in {
1293 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1294 (ins VR128:$src1, VR128:$src2),
1295 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1297 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1299 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1300 (ins VR128:$src1, VR128:$src2),
1301 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1303 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1306 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1307 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1308 (ins VR128:$src1, VR128:$src2),
1309 "movlhps\t{$src2, $dst|$dst, $src2}",
1311 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1312 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1313 (ins VR128:$src1, VR128:$src2),
1314 "movhlps\t{$src2, $dst|$dst, $src2}",
1316 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1319 let Predicates = [HasAVX] in {
1321 let AddedComplexity = 20 in {
1322 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1323 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1324 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1325 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1327 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1328 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1329 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1331 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1332 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1333 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1334 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1335 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1336 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1339 let AddedComplexity = 20 in {
1340 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1341 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1342 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1344 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1345 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1347 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1348 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1351 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1352 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1353 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1354 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1357 let Predicates = [HasSSE1] in {
1359 let AddedComplexity = 20 in {
1360 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1361 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1362 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1363 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1365 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1366 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1367 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1369 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1370 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1371 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1372 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1373 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1374 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1377 let AddedComplexity = 20 in {
1378 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1379 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1380 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1382 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1383 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1384 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1385 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1386 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1389 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1390 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1391 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1392 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1395 //===----------------------------------------------------------------------===//
1396 // SSE 1 & 2 - Conversion Instructions
1397 //===----------------------------------------------------------------------===//
1399 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1400 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1402 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1403 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1404 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1405 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1408 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1409 X86MemOperand x86memop, string asm> {
1410 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1412 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1415 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1416 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1417 string asm, Domain d> {
1418 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1419 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1420 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1421 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1424 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1425 X86MemOperand x86memop, string asm> {
1426 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1427 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1429 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1430 (ins DstRC:$src1, x86memop:$src),
1431 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1434 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1435 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1437 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1438 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1440 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1441 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1443 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1444 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1445 VEX, VEX_W, VEX_LIG;
1447 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1448 // register, but the same isn't true when only using memory operands,
1449 // provide other assembly "l" and "q" forms to address this explicitly
1450 // where appropriate to do so.
1451 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1453 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1454 VEX_4V, VEX_W, VEX_LIG;
1455 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1457 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1459 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1460 VEX_4V, VEX_W, VEX_LIG;
1462 let Predicates = [HasAVX] in {
1463 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1464 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1466 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1467 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1468 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1469 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1470 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1472 def : Pat<(f32 (sint_to_fp GR32:$src)),
1473 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1474 def : Pat<(f32 (sint_to_fp GR64:$src)),
1475 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1476 def : Pat<(f64 (sint_to_fp GR32:$src)),
1477 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1478 def : Pat<(f64 (sint_to_fp GR64:$src)),
1479 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1482 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1483 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1484 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1485 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1486 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1487 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1488 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1489 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1490 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1491 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1492 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1493 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1494 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1495 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1496 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1497 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1499 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1500 // and/or XMM operand(s).
1502 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1503 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1505 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1506 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1507 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1508 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1509 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1510 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1513 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1514 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1515 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1516 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1518 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1519 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1520 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1521 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1522 (ins DstRC:$src1, x86memop:$src2),
1524 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1525 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1526 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1529 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1530 f128mem, load, "cvtsd2si">, XD, VEX;
1531 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1532 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1535 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1536 // Get rid of this hack or rename the intrinsics, there are several
1537 // intructions that only match with the intrinsic form, why create duplicates
1538 // to let them be recognized by the assembler?
1539 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1540 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1541 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1542 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1545 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1546 f128mem, load, "cvtsd2si{l}">, XD;
1547 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1548 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1551 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1552 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1553 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1554 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1556 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1557 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1558 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1559 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1562 let Constraints = "$src1 = $dst" in {
1563 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1564 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1566 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1567 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1568 "cvtsi2ss{q}">, XS, REX_W;
1569 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1570 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1572 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1573 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1574 "cvtsi2sd">, XD, REX_W;
1579 // Aliases for intrinsics
1580 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1581 f32mem, load, "cvttss2si">, XS, VEX;
1582 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1583 int_x86_sse_cvttss2si64, f32mem, load,
1584 "cvttss2si">, XS, VEX, VEX_W;
1585 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1586 f128mem, load, "cvttsd2si">, XD, VEX;
1587 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1588 int_x86_sse2_cvttsd2si64, f128mem, load,
1589 "cvttsd2si">, XD, VEX, VEX_W;
1590 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1591 f32mem, load, "cvttss2si">, XS;
1592 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1593 int_x86_sse_cvttss2si64, f32mem, load,
1594 "cvttss2si{q}">, XS, REX_W;
1595 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1596 f128mem, load, "cvttsd2si">, XD;
1597 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1598 int_x86_sse2_cvttsd2si64, f128mem, load,
1599 "cvttsd2si{q}">, XD, REX_W;
1601 let Pattern = []<dag> in {
1602 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1603 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1605 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1606 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1608 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1609 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1610 SSEPackedSingle>, TB, VEX;
1611 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1612 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1613 SSEPackedSingle>, TB, VEX;
1616 let Pattern = []<dag> in {
1617 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1618 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1619 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1620 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1621 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1622 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1623 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1626 let Predicates = [HasSSE1] in {
1627 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1628 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1629 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1630 (CVTSS2SIrm addr:$src)>;
1631 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1632 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1633 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1634 (CVTSS2SI64rm addr:$src)>;
1637 let Predicates = [HasAVX] in {
1638 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1639 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1640 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1641 (VCVTSS2SIrm addr:$src)>;
1642 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1643 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1644 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1645 (VCVTSS2SI64rm addr:$src)>;
1650 // Convert scalar double to scalar single
1651 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1652 (ins FR64:$src1, FR64:$src2),
1653 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1656 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1657 (ins FR64:$src1, f64mem:$src2),
1658 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1659 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1661 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1664 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1665 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1666 [(set FR32:$dst, (fround FR64:$src))]>;
1667 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1668 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1669 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1670 Requires<[HasSSE2, OptForSize]>;
1672 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1673 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1675 let Constraints = "$src1 = $dst" in
1676 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1677 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1679 // Convert scalar single to scalar double
1680 // SSE2 instructions with XS prefix
1681 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1682 (ins FR32:$src1, FR32:$src2),
1683 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1684 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1686 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1687 (ins FR32:$src1, f32mem:$src2),
1688 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1689 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1691 let Predicates = [HasAVX] in {
1692 def : Pat<(f64 (fextend FR32:$src)),
1693 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1694 def : Pat<(fextend (loadf32 addr:$src)),
1695 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1696 def : Pat<(extloadf32 addr:$src),
1697 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1700 def : Pat<(extloadf32 addr:$src),
1701 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1702 Requires<[HasAVX, OptForSpeed]>;
1704 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1705 "cvtss2sd\t{$src, $dst|$dst, $src}",
1706 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1707 Requires<[HasSSE2]>;
1708 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1709 "cvtss2sd\t{$src, $dst|$dst, $src}",
1710 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1711 Requires<[HasSSE2, OptForSize]>;
1713 // extload f32 -> f64. This matches load+fextend because we have a hack in
1714 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1716 // Since these loads aren't folded into the fextend, we have to match it
1718 def : Pat<(fextend (loadf32 addr:$src)),
1719 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1720 def : Pat<(extloadf32 addr:$src),
1721 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1723 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1724 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1725 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1726 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1727 VR128:$src2))]>, XS, VEX_4V,
1729 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1730 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1731 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1732 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1733 (load addr:$src2)))]>, XS, VEX_4V,
1735 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1736 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1737 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1738 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1739 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1740 VR128:$src2))]>, XS,
1741 Requires<[HasSSE2]>;
1742 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1743 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1744 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1745 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1746 (load addr:$src2)))]>, XS,
1747 Requires<[HasSSE2]>;
1750 // Convert doubleword to packed single/double fp
1751 // SSE2 instructions without OpSize prefix
1752 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1753 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1754 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1755 TB, VEX, Requires<[HasAVX]>;
1756 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1757 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1758 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1759 (bitconvert (memopv2i64 addr:$src))))]>,
1760 TB, VEX, Requires<[HasAVX]>;
1761 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1762 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1763 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1764 TB, Requires<[HasSSE2]>;
1765 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1766 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1767 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1768 (bitconvert (memopv2i64 addr:$src))))]>,
1769 TB, Requires<[HasSSE2]>;
1771 // FIXME: why the non-intrinsic version is described as SSE3?
1772 // SSE2 instructions with XS prefix
1773 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1774 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1775 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1776 XS, VEX, Requires<[HasAVX]>;
1777 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1778 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1779 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1780 (bitconvert (memopv2i64 addr:$src))))]>,
1781 XS, VEX, Requires<[HasAVX]>;
1782 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1783 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1784 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1785 XS, Requires<[HasSSE2]>;
1786 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1787 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1788 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1789 (bitconvert (memopv2i64 addr:$src))))]>,
1790 XS, Requires<[HasSSE2]>;
1793 // Convert packed single/double fp to doubleword
1794 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1796 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1798 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1800 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1802 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1803 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1804 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1805 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1807 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1808 "cvtps2dq\t{$src, $dst|$dst, $src}",
1809 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1811 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1813 "cvtps2dq\t{$src, $dst|$dst, $src}",
1814 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1815 (memop addr:$src)))]>, VEX;
1816 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1817 "cvtps2dq\t{$src, $dst|$dst, $src}",
1818 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1819 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1820 "cvtps2dq\t{$src, $dst|$dst, $src}",
1821 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1822 (memop addr:$src)))]>;
1824 // SSE2 packed instructions with XD prefix
1825 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1826 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1827 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1828 XD, VEX, Requires<[HasAVX]>;
1829 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1830 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1831 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1832 (memop addr:$src)))]>,
1833 XD, VEX, Requires<[HasAVX]>;
1834 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1835 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1836 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1837 XD, Requires<[HasSSE2]>;
1838 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1839 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1840 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1841 (memop addr:$src)))]>,
1842 XD, Requires<[HasSSE2]>;
1845 // Convert with truncation packed single/double fp to doubleword
1846 // SSE2 packed instructions with XS prefix
1847 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1848 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1850 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1851 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1852 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1853 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1855 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1856 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1857 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1858 "cvttps2dq\t{$src, $dst|$dst, $src}",
1860 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1861 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1862 "cvttps2dq\t{$src, $dst|$dst, $src}",
1864 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1866 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1867 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1869 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1870 XS, VEX, Requires<[HasAVX]>;
1871 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1872 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1873 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1874 (memop addr:$src)))]>,
1875 XS, VEX, Requires<[HasAVX]>;
1877 let Predicates = [HasSSE2] in {
1878 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1879 (Int_CVTDQ2PSrr VR128:$src)>;
1880 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1881 (CVTTPS2DQrr VR128:$src)>;
1884 let Predicates = [HasAVX] in {
1885 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1886 (Int_VCVTDQ2PSrr VR128:$src)>;
1887 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1888 (VCVTTPS2DQrr VR128:$src)>;
1889 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1890 (VCVTDQ2PSYrr VR256:$src)>;
1891 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1892 (VCVTTPS2DQYrr VR256:$src)>;
1895 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1896 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1898 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1899 let isCodeGenOnly = 1 in
1900 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1901 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1902 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1903 (memop addr:$src)))]>, VEX;
1904 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1905 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1906 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1907 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1908 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1909 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1910 (memop addr:$src)))]>;
1912 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1913 // register, but the same isn't true when using memory operands instead.
1914 // Provide other assembly rr and rm forms to address this explicitly.
1915 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1916 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1919 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1920 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1921 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1922 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1925 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1926 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1927 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1928 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1930 // Convert packed single to packed double
1931 let Predicates = [HasAVX] in {
1932 // SSE2 instructions without OpSize prefix
1933 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1934 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1935 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1936 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1937 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1938 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1939 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1940 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1942 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1943 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1944 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1945 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1947 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1948 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1949 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1950 TB, VEX, Requires<[HasAVX]>;
1951 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1952 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1953 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1954 (load addr:$src)))]>,
1955 TB, VEX, Requires<[HasAVX]>;
1956 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1957 "cvtps2pd\t{$src, $dst|$dst, $src}",
1958 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1959 TB, Requires<[HasSSE2]>;
1960 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1961 "cvtps2pd\t{$src, $dst|$dst, $src}",
1962 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1963 (load addr:$src)))]>,
1964 TB, Requires<[HasSSE2]>;
1966 // Convert packed double to packed single
1967 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1968 // register, but the same isn't true when using memory operands instead.
1969 // Provide other assembly rr and rm forms to address this explicitly.
1970 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1971 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1972 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1973 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1976 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1977 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1978 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1979 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1982 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1983 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1984 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1985 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1986 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1987 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1988 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1989 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1992 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1993 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1994 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1995 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1997 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1998 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1999 (memop addr:$src)))]>;
2000 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2001 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2002 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2003 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2004 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2005 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2006 (memop addr:$src)))]>;
2008 // AVX 256-bit register conversion intrinsics
2009 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2010 // whenever possible to avoid declaring two versions of each one.
2011 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2012 (VCVTDQ2PSYrr VR256:$src)>;
2013 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2014 (VCVTDQ2PSYrm addr:$src)>;
2016 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2017 (VCVTPD2PSYrr VR256:$src)>;
2018 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2019 (VCVTPD2PSYrm addr:$src)>;
2021 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2022 (VCVTPS2DQYrr VR256:$src)>;
2023 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2024 (VCVTPS2DQYrm addr:$src)>;
2026 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2027 (VCVTPS2PDYrr VR128:$src)>;
2028 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2029 (VCVTPS2PDYrm addr:$src)>;
2031 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2032 (VCVTTPD2DQYrr VR256:$src)>;
2033 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2034 (VCVTTPD2DQYrm addr:$src)>;
2036 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2037 (VCVTTPS2DQYrr VR256:$src)>;
2038 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2039 (VCVTTPS2DQYrm addr:$src)>;
2041 // Match fround and fextend for 128/256-bit conversions
2042 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2043 (VCVTPD2PSYrr VR256:$src)>;
2044 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2045 (VCVTPD2PSYrm addr:$src)>;
2047 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2048 (VCVTPS2PDYrr VR128:$src)>;
2049 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2050 (VCVTPS2PDYrm addr:$src)>;
2052 //===----------------------------------------------------------------------===//
2053 // SSE 1 & 2 - Compare Instructions
2054 //===----------------------------------------------------------------------===//
2056 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2057 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2058 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2059 string asm, string asm_alt> {
2060 def rr : SIi8<0xC2, MRMSrcReg,
2061 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2062 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2063 def rm : SIi8<0xC2, MRMSrcMem,
2064 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2065 [(set RC:$dst, (OpNode (VT RC:$src1),
2066 (ld_frag addr:$src2), imm:$cc))]>;
2068 // Accept explicit immediate argument form instead of comparison code.
2069 let neverHasSideEffects = 1 in {
2070 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2071 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2073 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2074 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2078 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2079 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2080 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2081 XS, VEX_4V, VEX_LIG;
2082 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2083 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2084 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2085 XD, VEX_4V, VEX_LIG;
2087 let Constraints = "$src1 = $dst" in {
2088 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2089 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2090 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2092 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2093 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2094 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2098 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2099 Intrinsic Int, string asm> {
2100 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2101 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2102 [(set VR128:$dst, (Int VR128:$src1,
2103 VR128:$src, imm:$cc))]>;
2104 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2105 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2106 [(set VR128:$dst, (Int VR128:$src1,
2107 (load addr:$src), imm:$cc))]>;
2110 // Aliases to match intrinsics which expect XMM operand(s).
2111 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2112 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2114 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2115 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2117 let Constraints = "$src1 = $dst" in {
2118 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2119 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2120 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2121 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2125 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2126 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2127 ValueType vt, X86MemOperand x86memop,
2128 PatFrag ld_frag, string OpcodeStr, Domain d> {
2129 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2130 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2131 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2132 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2133 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2134 [(set EFLAGS, (OpNode (vt RC:$src1),
2135 (ld_frag addr:$src2)))], d>;
2138 let Defs = [EFLAGS] in {
2139 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2140 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2141 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2142 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2144 let Pattern = []<dag> in {
2145 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2146 "comiss", SSEPackedSingle>, TB, VEX,
2148 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2149 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2153 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2154 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2155 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2156 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2158 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2159 load, "comiss", SSEPackedSingle>, TB, VEX;
2160 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2161 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2162 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2163 "ucomiss", SSEPackedSingle>, TB;
2164 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2165 "ucomisd", SSEPackedDouble>, TB, OpSize;
2167 let Pattern = []<dag> in {
2168 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2169 "comiss", SSEPackedSingle>, TB;
2170 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2171 "comisd", SSEPackedDouble>, TB, OpSize;
2174 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2175 load, "ucomiss", SSEPackedSingle>, TB;
2176 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2177 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2179 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2180 "comiss", SSEPackedSingle>, TB;
2181 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2182 "comisd", SSEPackedDouble>, TB, OpSize;
2183 } // Defs = [EFLAGS]
2185 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2186 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2187 Intrinsic Int, string asm, string asm_alt,
2189 let isAsmParserOnly = 1 in {
2190 def rri : PIi8<0xC2, MRMSrcReg,
2191 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2192 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2193 def rmi : PIi8<0xC2, MRMSrcMem,
2194 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2195 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2198 // Accept explicit immediate argument form instead of comparison code.
2199 def rri_alt : PIi8<0xC2, MRMSrcReg,
2200 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2202 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2203 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2207 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2208 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2209 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2210 SSEPackedSingle>, TB, VEX_4V;
2211 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2212 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2213 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2214 SSEPackedDouble>, TB, OpSize, VEX_4V;
2215 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2216 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2217 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2218 SSEPackedSingle>, TB, VEX_4V;
2219 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2220 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2221 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2222 SSEPackedDouble>, TB, OpSize, VEX_4V;
2223 let Constraints = "$src1 = $dst" in {
2224 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2225 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2226 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2227 SSEPackedSingle>, TB;
2228 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2229 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2230 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2231 SSEPackedDouble>, TB, OpSize;
2234 let Predicates = [HasAVX] in {
2235 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2236 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2237 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2238 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2239 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2240 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2241 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2242 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2244 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2245 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2246 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2247 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2248 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2249 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2250 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2251 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2254 let Predicates = [HasSSE1] in {
2255 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2256 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2257 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2258 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2261 let Predicates = [HasSSE2] in {
2262 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2263 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2264 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2265 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2268 //===----------------------------------------------------------------------===//
2269 // SSE 1 & 2 - Shuffle Instructions
2270 //===----------------------------------------------------------------------===//
2272 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2273 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2274 ValueType vt, string asm, PatFrag mem_frag,
2275 Domain d, bit IsConvertibleToThreeAddress = 0> {
2276 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2277 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2278 [(set RC:$dst, (vt (shufp:$src3
2279 RC:$src1, (mem_frag addr:$src2))))], d>;
2280 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2281 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2282 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2284 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2287 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2288 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2289 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2290 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2291 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2292 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2293 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2294 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2295 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2296 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2297 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2298 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2300 let Constraints = "$src1 = $dst" in {
2301 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2302 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2303 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2305 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2306 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2307 memopv2f64, SSEPackedDouble>, TB, OpSize;
2310 let Predicates = [HasAVX] in {
2311 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2312 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2313 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2314 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2315 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2316 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2317 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2318 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2319 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2320 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2321 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2322 // fall back to this for SSE1)
2323 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2324 (VSHUFPSrri VR128:$src2, VR128:$src1,
2325 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2326 // Special unary SHUFPSrri case.
2327 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2328 (VSHUFPSrri VR128:$src1, VR128:$src1,
2329 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 // Special binary v4i32 shuffle cases with SHUFPS.
2331 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2332 (VSHUFPSrri VR128:$src1, VR128:$src2,
2333 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2334 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2335 (bc_v4i32 (memopv2i64 addr:$src2)))),
2336 (VSHUFPSrmi VR128:$src1, addr:$src2,
2337 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2338 // Special unary SHUFPDrri cases.
2339 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2340 (VSHUFPDrri VR128:$src1, VR128:$src1,
2341 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2342 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2343 (VSHUFPDrri VR128:$src1, VR128:$src1,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2345 // Special binary v2i64 shuffle cases using SHUFPDrri.
2346 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2347 (VSHUFPDrri VR128:$src1, VR128:$src2,
2348 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2350 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2351 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2352 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2353 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2354 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2355 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2356 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2357 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2358 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2359 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2362 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2363 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2364 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2365 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2366 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2368 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2369 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2370 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2371 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2372 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2374 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2375 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2376 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2377 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2378 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2380 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2381 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2382 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2383 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2384 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2387 let Predicates = [HasSSE1] in {
2388 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2389 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2390 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2391 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2392 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2393 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2394 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2395 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2396 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2397 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2398 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2399 // fall back to this for SSE1)
2400 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2401 (SHUFPSrri VR128:$src2, VR128:$src1,
2402 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2403 // Special unary SHUFPSrri case.
2404 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2405 (SHUFPSrri VR128:$src1, VR128:$src1,
2406 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2409 let Predicates = [HasSSE2] in {
2410 // Special binary v4i32 shuffle cases with SHUFPS.
2411 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2412 (SHUFPSrri VR128:$src1, VR128:$src2,
2413 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2414 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2415 (bc_v4i32 (memopv2i64 addr:$src2)))),
2416 (SHUFPSrmi VR128:$src1, addr:$src2,
2417 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2418 // Special unary SHUFPDrri cases.
2419 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2420 (SHUFPDrri VR128:$src1, VR128:$src1,
2421 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2422 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2423 (SHUFPDrri VR128:$src1, VR128:$src1,
2424 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2425 // Special binary v2i64 shuffle cases using SHUFPDrri.
2426 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2427 (SHUFPDrri VR128:$src1, VR128:$src2,
2428 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2429 // Generic SHUFPD patterns
2430 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2431 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2432 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2433 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2434 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2435 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2436 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2437 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2438 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2439 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2442 //===----------------------------------------------------------------------===//
2443 // SSE 1 & 2 - Unpack Instructions
2444 //===----------------------------------------------------------------------===//
2446 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2447 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2448 PatFrag mem_frag, RegisterClass RC,
2449 X86MemOperand x86memop, string asm,
2451 def rr : PI<opc, MRMSrcReg,
2452 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2454 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2455 def rm : PI<opc, MRMSrcMem,
2456 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2458 (vt (OpNode RC:$src1,
2459 (mem_frag addr:$src2))))], d>;
2462 let AddedComplexity = 10 in {
2463 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2464 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2465 SSEPackedSingle>, TB, VEX_4V;
2466 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2467 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2468 SSEPackedDouble>, TB, OpSize, VEX_4V;
2469 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2470 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2471 SSEPackedSingle>, TB, VEX_4V;
2472 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2473 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2474 SSEPackedDouble>, TB, OpSize, VEX_4V;
2476 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2477 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2478 SSEPackedSingle>, TB, VEX_4V;
2479 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2480 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2481 SSEPackedDouble>, TB, OpSize, VEX_4V;
2482 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2483 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2484 SSEPackedSingle>, TB, VEX_4V;
2485 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2486 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2487 SSEPackedDouble>, TB, OpSize, VEX_4V;
2489 let Constraints = "$src1 = $dst" in {
2490 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2491 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2492 SSEPackedSingle>, TB;
2493 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2494 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2495 SSEPackedDouble>, TB, OpSize;
2496 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2497 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2498 SSEPackedSingle>, TB;
2499 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2500 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2501 SSEPackedDouble>, TB, OpSize;
2502 } // Constraints = "$src1 = $dst"
2503 } // AddedComplexity
2505 let Predicates = [HasSSE1] in {
2506 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2507 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2508 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2509 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2510 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2511 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2512 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2513 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2516 let Predicates = [HasSSE2] in {
2517 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2518 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2519 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2520 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2521 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2522 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2523 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2524 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2526 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2527 // problem is during lowering, where it's not possible to recognize the load
2528 // fold cause it has two uses through a bitcast. One use disappears at isel
2529 // time and the fold opportunity reappears.
2530 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2531 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2533 let AddedComplexity = 10 in
2534 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2535 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2538 let Predicates = [HasAVX] in {
2539 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2540 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2541 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2542 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2543 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2544 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2545 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2546 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2548 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2549 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2550 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2551 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2552 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2553 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2554 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2555 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2557 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2558 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2559 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2560 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2561 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2562 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2563 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2564 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2566 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2567 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2568 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2569 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2570 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2571 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2572 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2573 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2575 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2576 // problem is during lowering, where it's not possible to recognize the load
2577 // fold cause it has two uses through a bitcast. One use disappears at isel
2578 // time and the fold opportunity reappears.
2579 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2580 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2581 let AddedComplexity = 10 in
2582 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2583 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2586 //===----------------------------------------------------------------------===//
2587 // SSE 1 & 2 - Extract Floating-Point Sign mask
2588 //===----------------------------------------------------------------------===//
2590 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2591 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2593 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2594 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2595 [(set GR32:$dst, (Int RC:$src))], d>;
2596 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2597 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2600 let Predicates = [HasAVX] in {
2601 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2602 "movmskps", SSEPackedSingle>, TB, VEX;
2603 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2604 "movmskpd", SSEPackedDouble>, TB,
2606 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2607 "movmskps", SSEPackedSingle>, TB, VEX;
2608 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2609 "movmskpd", SSEPackedDouble>, TB,
2612 def : Pat<(i32 (X86fgetsign FR32:$src)),
2613 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2615 def : Pat<(i64 (X86fgetsign FR32:$src)),
2616 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2618 def : Pat<(i32 (X86fgetsign FR64:$src)),
2619 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2621 def : Pat<(i64 (X86fgetsign FR64:$src)),
2622 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2626 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2627 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2628 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2629 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2631 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2632 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2633 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2634 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2638 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2639 SSEPackedSingle>, TB;
2640 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2641 SSEPackedDouble>, TB, OpSize;
2643 def : Pat<(i32 (X86fgetsign FR32:$src)),
2644 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2645 sub_ss))>, Requires<[HasSSE1]>;
2646 def : Pat<(i64 (X86fgetsign FR32:$src)),
2647 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2648 sub_ss))>, Requires<[HasSSE1]>;
2649 def : Pat<(i32 (X86fgetsign FR64:$src)),
2650 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2651 sub_sd))>, Requires<[HasSSE2]>;
2652 def : Pat<(i64 (X86fgetsign FR64:$src)),
2653 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2654 sub_sd))>, Requires<[HasSSE2]>;
2656 //===---------------------------------------------------------------------===//
2657 // SSE2 - Packed Integer Logical Instructions
2658 //===---------------------------------------------------------------------===//
2660 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2662 /// PDI_binop_rm - Simple SSE2 binary operator.
2663 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2664 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2665 X86MemOperand x86memop, bit IsCommutable = 0,
2667 let isCommutable = IsCommutable in
2668 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2669 (ins RC:$src1, RC:$src2),
2671 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2672 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2673 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2674 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2675 (ins RC:$src1, x86memop:$src2),
2677 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2678 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2679 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2680 (bitconvert (memop_frag addr:$src2)))))]>;
2682 } // ExeDomain = SSEPackedInt
2684 // These are ordered here for pattern ordering requirements with the fp versions
2686 let Predicates = [HasAVX] in {
2687 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2688 i128mem, 1, 0>, VEX_4V;
2689 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2690 i128mem, 1, 0>, VEX_4V;
2691 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2692 i128mem, 1, 0>, VEX_4V;
2693 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2694 i128mem, 0, 0>, VEX_4V;
2697 let Constraints = "$src1 = $dst" in {
2698 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2700 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2702 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2704 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2706 } // Constraints = "$src1 = $dst"
2708 let Predicates = [HasAVX2] in {
2709 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2710 i256mem, 1, 0>, VEX_4V;
2711 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2712 i256mem, 1, 0>, VEX_4V;
2713 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2714 i256mem, 1, 0>, VEX_4V;
2715 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2716 i256mem, 0, 0>, VEX_4V;
2719 //===----------------------------------------------------------------------===//
2720 // SSE 1 & 2 - Logical Instructions
2721 //===----------------------------------------------------------------------===//
2723 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2725 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2727 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2728 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2730 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2731 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2733 let Constraints = "$src1 = $dst" in {
2734 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2735 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2737 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2738 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2742 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2743 let mayLoad = 0 in {
2744 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2745 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2746 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2749 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2750 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2752 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2754 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2756 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2757 // are all promoted to v2i64, and the patterns are covered by the int
2758 // version. This is needed in SSE only, because v2i64 isn't supported on
2759 // SSE1, but only on SSE2.
2760 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2761 !strconcat(OpcodeStr, "ps"), f128mem, [],
2762 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2763 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2765 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2766 !strconcat(OpcodeStr, "pd"), f128mem,
2767 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2768 (bc_v2i64 (v2f64 VR128:$src2))))],
2769 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2770 (memopv2i64 addr:$src2)))], 0>,
2772 let Constraints = "$src1 = $dst" in {
2773 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2774 !strconcat(OpcodeStr, "ps"), f128mem,
2775 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2776 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2777 (memopv2i64 addr:$src2)))]>, TB;
2779 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2780 !strconcat(OpcodeStr, "pd"), f128mem,
2781 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2782 (bc_v2i64 (v2f64 VR128:$src2))))],
2783 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2784 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2788 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2790 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2792 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2793 !strconcat(OpcodeStr, "ps"), f256mem,
2794 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2795 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2796 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2798 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2799 !strconcat(OpcodeStr, "pd"), f256mem,
2800 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2801 (bc_v4i64 (v4f64 VR256:$src2))))],
2802 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2803 (memopv4i64 addr:$src2)))], 0>,
2807 // AVX 256-bit packed logical ops forms
2808 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2809 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2810 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2811 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2813 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2814 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2815 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2816 let isCommutable = 0 in
2817 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2819 //===----------------------------------------------------------------------===//
2820 // SSE 1 & 2 - Arithmetic Instructions
2821 //===----------------------------------------------------------------------===//
2823 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2826 /// In addition, we also have a special variant of the scalar form here to
2827 /// represent the associated intrinsic operation. This form is unlike the
2828 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2829 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2831 /// These three forms can each be reg+reg or reg+mem.
2834 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2836 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2838 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2839 OpNode, FR32, f32mem, Is2Addr>, XS;
2840 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2841 OpNode, FR64, f64mem, Is2Addr>, XD;
2844 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2846 let mayLoad = 0 in {
2847 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2848 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2849 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2850 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2854 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2856 let mayLoad = 0 in {
2857 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2858 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2859 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2860 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2864 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2866 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2867 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2868 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2869 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2872 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2874 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2875 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2876 SSEPackedSingle, Is2Addr>, TB;
2878 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2879 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2880 SSEPackedDouble, Is2Addr>, TB, OpSize;
2883 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2884 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2885 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2886 SSEPackedSingle, 0>, TB;
2888 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2889 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2890 SSEPackedDouble, 0>, TB, OpSize;
2893 // Binary Arithmetic instructions
2894 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2895 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2896 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2897 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2898 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2899 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2900 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2901 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2903 let isCommutable = 0 in {
2904 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2905 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2906 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2907 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2908 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2909 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2910 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2911 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2912 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2913 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2914 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2915 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2916 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2917 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2918 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2919 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2920 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2921 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2922 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2923 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2926 let Constraints = "$src1 = $dst" in {
2927 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2928 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2929 basic_sse12_fp_binop_s_int<0x58, "add">;
2930 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2931 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2932 basic_sse12_fp_binop_s_int<0x59, "mul">;
2934 let isCommutable = 0 in {
2935 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2936 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2937 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2938 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2939 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2940 basic_sse12_fp_binop_s_int<0x5E, "div">;
2941 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2942 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2943 basic_sse12_fp_binop_s_int<0x5F, "max">,
2944 basic_sse12_fp_binop_p_int<0x5F, "max">;
2945 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2946 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2947 basic_sse12_fp_binop_s_int<0x5D, "min">,
2948 basic_sse12_fp_binop_p_int<0x5D, "min">;
2953 /// In addition, we also have a special variant of the scalar form here to
2954 /// represent the associated intrinsic operation. This form is unlike the
2955 /// plain scalar form, in that it takes an entire vector (instead of a
2956 /// scalar) and leaves the top elements undefined.
2958 /// And, we have a special variant form for a full-vector intrinsic form.
2960 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2961 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2962 SDNode OpNode, Intrinsic F32Int> {
2963 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2964 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2965 [(set FR32:$dst, (OpNode FR32:$src))]>;
2966 // For scalar unary operations, fold a load into the operation
2967 // only in OptForSize mode. It eliminates an instruction, but it also
2968 // eliminates a whole-register clobber (the load), so it introduces a
2969 // partial register update condition.
2970 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2971 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2972 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2973 Requires<[HasSSE1, OptForSize]>;
2974 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2975 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2976 [(set VR128:$dst, (F32Int VR128:$src))]>;
2977 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2978 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2982 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2983 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2984 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2985 !strconcat(OpcodeStr,
2986 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2988 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2989 !strconcat(OpcodeStr,
2990 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2991 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2992 (ins VR128:$src1, ssmem:$src2),
2993 !strconcat(OpcodeStr,
2994 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2997 /// sse1_fp_unop_p - SSE1 unops in packed form.
2998 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2999 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3000 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3001 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3002 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3003 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3004 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3007 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3008 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3009 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3010 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3011 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3012 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3013 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3014 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3017 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3018 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3019 Intrinsic V4F32Int> {
3020 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3021 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3022 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3023 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3024 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3025 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3028 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3029 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3030 Intrinsic V4F32Int> {
3031 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3032 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3033 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3034 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3035 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3036 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3039 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3040 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3041 SDNode OpNode, Intrinsic F64Int> {
3042 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3043 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3044 [(set FR64:$dst, (OpNode FR64:$src))]>;
3045 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3046 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3047 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3048 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3049 Requires<[HasSSE2, OptForSize]>;
3050 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3051 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3052 [(set VR128:$dst, (F64Int VR128:$src))]>;
3053 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3054 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3055 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3058 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3059 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3060 let neverHasSideEffects = 1 in {
3061 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3062 !strconcat(OpcodeStr,
3063 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3065 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3066 !strconcat(OpcodeStr,
3067 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3069 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3070 (ins VR128:$src1, sdmem:$src2),
3071 !strconcat(OpcodeStr,
3072 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3075 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3076 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3078 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3079 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3080 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3081 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3082 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3083 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3086 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3087 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3088 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3089 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3090 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3091 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3092 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3093 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3096 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3097 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3098 Intrinsic V2F64Int> {
3099 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3100 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3101 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3102 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3103 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3104 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3107 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3108 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3109 Intrinsic V2F64Int> {
3110 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3111 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3112 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3113 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3114 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3115 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3118 let Predicates = [HasAVX] in {
3120 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3121 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3123 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3124 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3125 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3126 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3127 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3128 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3129 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3130 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3133 // Reciprocal approximations. Note that these typically require refinement
3134 // in order to obtain suitable precision.
3135 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3136 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3137 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3138 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3139 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3141 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3142 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3143 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3144 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3145 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3148 def : Pat<(f32 (fsqrt FR32:$src)),
3149 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3150 def : Pat<(f32 (fsqrt (load addr:$src))),
3151 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3152 Requires<[HasAVX, OptForSize]>;
3153 def : Pat<(f64 (fsqrt FR64:$src)),
3154 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3155 def : Pat<(f64 (fsqrt (load addr:$src))),
3156 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3157 Requires<[HasAVX, OptForSize]>;
3159 def : Pat<(f32 (X86frsqrt FR32:$src)),
3160 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3161 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3162 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3163 Requires<[HasAVX, OptForSize]>;
3165 def : Pat<(f32 (X86frcp FR32:$src)),
3166 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3167 def : Pat<(f32 (X86frcp (load addr:$src))),
3168 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3169 Requires<[HasAVX, OptForSize]>;
3171 let Predicates = [HasAVX] in {
3172 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3173 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3174 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3175 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3177 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3178 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3180 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3181 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3182 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3183 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3185 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3186 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3188 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3189 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3190 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3191 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3193 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3194 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3196 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3197 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3198 (VRCPSSr (f32 (IMPLICIT_DEF)),
3199 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3201 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3202 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3206 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3207 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3208 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3209 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3210 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3211 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3213 // Reciprocal approximations. Note that these typically require refinement
3214 // in order to obtain suitable precision.
3215 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3216 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3217 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3218 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3219 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3220 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3222 // There is no f64 version of the reciprocal approximation instructions.
3224 //===----------------------------------------------------------------------===//
3225 // SSE 1 & 2 - Non-temporal stores
3226 //===----------------------------------------------------------------------===//
3228 let AddedComplexity = 400 in { // Prefer non-temporal versions
3229 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3230 (ins f128mem:$dst, VR128:$src),
3231 "movntps\t{$src, $dst|$dst, $src}",
3232 [(alignednontemporalstore (v4f32 VR128:$src),
3234 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3235 (ins f128mem:$dst, VR128:$src),
3236 "movntpd\t{$src, $dst|$dst, $src}",
3237 [(alignednontemporalstore (v2f64 VR128:$src),
3240 let ExeDomain = SSEPackedInt in
3241 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3242 (ins f128mem:$dst, VR128:$src),
3243 "movntdq\t{$src, $dst|$dst, $src}",
3244 [(alignednontemporalstore (v2i64 VR128:$src),
3247 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3248 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3250 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3251 (ins f256mem:$dst, VR256:$src),
3252 "movntps\t{$src, $dst|$dst, $src}",
3253 [(alignednontemporalstore (v8f32 VR256:$src),
3255 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3256 (ins f256mem:$dst, VR256:$src),
3257 "movntpd\t{$src, $dst|$dst, $src}",
3258 [(alignednontemporalstore (v4f64 VR256:$src),
3260 let ExeDomain = SSEPackedInt in
3261 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3262 (ins f256mem:$dst, VR256:$src),
3263 "movntdq\t{$src, $dst|$dst, $src}",
3264 [(alignednontemporalstore (v4i64 VR256:$src),
3268 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3269 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3270 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3271 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3272 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3273 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3275 let AddedComplexity = 400 in { // Prefer non-temporal versions
3276 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3277 "movntps\t{$src, $dst|$dst, $src}",
3278 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3279 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3280 "movntpd\t{$src, $dst|$dst, $src}",
3281 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3283 let ExeDomain = SSEPackedInt in
3284 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3285 "movntdq\t{$src, $dst|$dst, $src}",
3286 [(alignednontemporalstore (v2i64 VR128:$src), addr:$dst)]>;
3288 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3289 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3291 // There is no AVX form for instructions below this point
3292 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3293 "movnti{l}\t{$src, $dst|$dst, $src}",
3294 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3295 TB, Requires<[HasXMMInt]>;
3296 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3297 "movnti{q}\t{$src, $dst|$dst, $src}",
3298 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3299 TB, Requires<[HasXMMInt]>;
3302 //===----------------------------------------------------------------------===//
3303 // SSE 1 & 2 - Prefetch and memory fence
3304 //===----------------------------------------------------------------------===//
3306 // Prefetch intrinsic.
3307 let Predicates = [HasXMM] in {
3308 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3309 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3310 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3311 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3312 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3313 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3314 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3315 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3319 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3320 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3321 TB, Requires<[HasXMMInt]>;
3323 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3324 // was introduced with SSE2, it's backward compatible.
3325 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3327 // Load, store, and memory fence
3328 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3329 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasXMM]>;
3330 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3331 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasXMMInt]>;
3332 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3333 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasXMMInt]>;
3335 def : Pat<(X86SFence), (SFENCE)>;
3336 def : Pat<(X86LFence), (LFENCE)>;
3337 def : Pat<(X86MFence), (MFENCE)>;
3339 //===----------------------------------------------------------------------===//
3340 // SSE 1 & 2 - Load/Store XCSR register
3341 //===----------------------------------------------------------------------===//
3343 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3344 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3345 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3346 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3348 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3349 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3350 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3351 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3353 //===---------------------------------------------------------------------===//
3354 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3355 //===---------------------------------------------------------------------===//
3357 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3359 let neverHasSideEffects = 1 in {
3360 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3361 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3362 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3363 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3365 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3366 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3367 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3368 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3371 let isCodeGenOnly = 1 in {
3372 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3373 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3374 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3375 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3376 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3377 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3378 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3379 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3382 let canFoldAsLoad = 1, mayLoad = 1 in {
3383 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3384 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3385 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3386 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3387 let Predicates = [HasAVX] in {
3388 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3389 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3390 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3391 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3395 let mayStore = 1 in {
3396 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3397 (ins i128mem:$dst, VR128:$src),
3398 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3399 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3400 (ins i256mem:$dst, VR256:$src),
3401 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3402 let Predicates = [HasAVX] in {
3403 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3404 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3405 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3406 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3410 let neverHasSideEffects = 1 in
3411 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3412 "movdqa\t{$src, $dst|$dst, $src}", []>;
3414 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3415 "movdqu\t{$src, $dst|$dst, $src}",
3416 []>, XS, Requires<[HasSSE2]>;
3419 let isCodeGenOnly = 1 in {
3420 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3421 "movdqa\t{$src, $dst|$dst, $src}", []>;
3423 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3424 "movdqu\t{$src, $dst|$dst, $src}",
3425 []>, XS, Requires<[HasSSE2]>;
3428 let canFoldAsLoad = 1, mayLoad = 1 in {
3429 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3430 "movdqa\t{$src, $dst|$dst, $src}",
3431 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3432 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3433 "movdqu\t{$src, $dst|$dst, $src}",
3434 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3435 XS, Requires<[HasSSE2]>;
3438 let mayStore = 1 in {
3439 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3440 "movdqa\t{$src, $dst|$dst, $src}",
3441 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3442 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3443 "movdqu\t{$src, $dst|$dst, $src}",
3444 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3445 XS, Requires<[HasSSE2]>;
3448 // Intrinsic forms of MOVDQU load and store
3449 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3450 "vmovdqu\t{$src, $dst|$dst, $src}",
3451 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3452 XS, VEX, Requires<[HasAVX]>;
3454 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3455 "movdqu\t{$src, $dst|$dst, $src}",
3456 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3457 XS, Requires<[HasSSE2]>;
3459 } // ExeDomain = SSEPackedInt
3461 let Predicates = [HasAVX] in {
3462 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3463 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3464 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3467 //===---------------------------------------------------------------------===//
3468 // SSE2 - Packed Integer Arithmetic Instructions
3469 //===---------------------------------------------------------------------===//
3471 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3473 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3474 RegisterClass RC, PatFrag memop_frag,
3475 X86MemOperand x86memop, bit IsCommutable = 0,
3477 let isCommutable = IsCommutable in
3478 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3479 (ins RC:$src1, RC:$src2),
3481 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3482 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3483 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3484 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3485 (ins RC:$src1, x86memop:$src2),
3487 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3488 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3489 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3492 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3493 string OpcodeStr, Intrinsic IntId,
3494 Intrinsic IntId2, RegisterClass RC,
3496 // src2 is always 128-bit
3497 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3498 (ins RC:$src1, VR128:$src2),
3500 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3501 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3502 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3503 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3504 (ins RC:$src1, i128mem:$src2),
3506 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3507 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3508 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3509 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3510 (ins RC:$src1, i32i8imm:$src2),
3512 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3513 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3514 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3517 } // ExeDomain = SSEPackedInt
3519 // 128-bit Integer Arithmetic
3521 let Predicates = [HasAVX] in {
3522 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3523 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3524 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3525 i128mem, 1, 0>, VEX_4V;
3526 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3527 i128mem, 1, 0>, VEX_4V;
3528 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3529 i128mem, 1, 0>, VEX_4V;
3530 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3531 i128mem, 1, 0>, VEX_4V;
3532 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3533 i128mem, 0, 0>, VEX_4V;
3534 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3535 i128mem, 0, 0>, VEX_4V;
3536 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3537 i128mem, 0, 0>, VEX_4V;
3538 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3539 i128mem, 0, 0>, VEX_4V;
3542 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3543 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3544 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3545 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3546 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3547 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3548 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3549 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3550 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3551 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3552 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3553 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3554 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3555 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3556 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3557 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3558 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3559 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3560 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3561 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3562 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3563 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3564 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3565 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3566 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3567 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3568 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3569 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3570 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3571 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3572 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3573 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3574 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3575 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3576 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3577 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3578 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3579 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3582 let Predicates = [HasAVX2] in {
3583 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3584 i256mem, 1, 0>, VEX_4V;
3585 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3586 i256mem, 1, 0>, VEX_4V;
3587 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3588 i256mem, 1, 0>, VEX_4V;
3589 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3590 i256mem, 1, 0>, VEX_4V;
3591 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3592 i256mem, 1, 0>, VEX_4V;
3593 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3594 i256mem, 0, 0>, VEX_4V;
3595 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3596 i256mem, 0, 0>, VEX_4V;
3597 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3598 i256mem, 0, 0>, VEX_4V;
3599 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3600 i256mem, 0, 0>, VEX_4V;
3603 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3604 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3605 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3606 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3607 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3608 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3609 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3610 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3611 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3612 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3613 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3614 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3615 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3616 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3617 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3618 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3619 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3620 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3621 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3622 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3623 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3624 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3625 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3626 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3627 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3628 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3629 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3630 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3631 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3632 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3633 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3634 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3635 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3636 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3637 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3638 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3639 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3640 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3643 let Constraints = "$src1 = $dst" in {
3644 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3646 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3648 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3650 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3652 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3654 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3656 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3658 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3660 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3664 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3665 VR128, memopv2i64, i128mem>;
3666 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3667 VR128, memopv2i64, i128mem>;
3668 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3669 VR128, memopv2i64, i128mem>;
3670 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3671 VR128, memopv2i64, i128mem>;
3672 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3673 VR128, memopv2i64, i128mem, 1>;
3674 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3675 VR128, memopv2i64, i128mem, 1>;
3676 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3677 VR128, memopv2i64, i128mem, 1>;
3678 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3679 VR128, memopv2i64, i128mem, 1>;
3680 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3681 VR128, memopv2i64, i128mem, 1>;
3682 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3683 VR128, memopv2i64, i128mem, 1>;
3684 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3685 VR128, memopv2i64, i128mem, 1>;
3686 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3687 VR128, memopv2i64, i128mem, 1>;
3688 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3689 VR128, memopv2i64, i128mem, 1>;
3690 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3691 VR128, memopv2i64, i128mem, 1>;
3692 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3693 VR128, memopv2i64, i128mem, 1>;
3694 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3695 VR128, memopv2i64, i128mem, 1>;
3696 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3697 VR128, memopv2i64, i128mem, 1>;
3698 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3699 VR128, memopv2i64, i128mem, 1>;
3700 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3701 VR128, memopv2i64, i128mem, 1>;
3703 } // Constraints = "$src1 = $dst"
3705 //===---------------------------------------------------------------------===//
3706 // SSE2 - Packed Integer Logical Instructions
3707 //===---------------------------------------------------------------------===//
3709 let Predicates = [HasAVX] in {
3710 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3711 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3713 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3714 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3716 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3717 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3720 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3721 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3723 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3724 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3726 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3727 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3730 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3731 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3733 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3734 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3737 let ExeDomain = SSEPackedInt in {
3738 let neverHasSideEffects = 1 in {
3739 // 128-bit logical shifts.
3740 def VPSLLDQri : PDIi8<0x73, MRM7r,
3741 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3742 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3744 def VPSRLDQri : PDIi8<0x73, MRM3r,
3745 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3746 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3748 // PSRADQri doesn't exist in SSE[1-3].
3753 let Predicates = [HasAVX2] in {
3754 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3755 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3757 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3758 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3760 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3761 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3764 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3765 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3767 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3768 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3770 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3771 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3774 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3775 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3777 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3778 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3781 let ExeDomain = SSEPackedInt in {
3782 let neverHasSideEffects = 1 in {
3783 // 128-bit logical shifts.
3784 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3785 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3786 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3788 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3789 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3790 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3792 // PSRADQYri doesn't exist in SSE[1-3].
3797 let Constraints = "$src1 = $dst" in {
3798 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3799 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3801 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3802 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3804 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3805 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3808 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3809 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3811 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3812 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3814 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3815 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3818 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3819 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3821 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3822 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3825 let ExeDomain = SSEPackedInt in {
3826 let neverHasSideEffects = 1 in {
3827 // 128-bit logical shifts.
3828 def PSLLDQri : PDIi8<0x73, MRM7r,
3829 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3830 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3831 def PSRLDQri : PDIi8<0x73, MRM3r,
3832 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3833 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3834 // PSRADQri doesn't exist in SSE[1-3].
3837 } // Constraints = "$src1 = $dst"
3839 let Predicates = [HasAVX] in {
3840 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3841 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3842 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3843 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3844 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3845 (VPSLLDQri VR128:$src1, imm:$src2)>;
3846 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3847 (VPSRLDQri VR128:$src1, imm:$src2)>;
3848 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3849 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3851 // Shift up / down and insert zero's.
3852 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3853 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3854 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3855 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3858 let Predicates = [HasAVX2] in {
3859 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3860 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3861 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3862 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3863 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3864 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3865 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3866 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3869 let Predicates = [HasSSE2] in {
3870 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3871 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3872 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3873 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3874 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3875 (PSLLDQri VR128:$src1, imm:$src2)>;
3876 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3877 (PSRLDQri VR128:$src1, imm:$src2)>;
3878 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3879 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3881 // Shift up / down and insert zero's.
3882 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3883 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3884 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3885 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3888 //===---------------------------------------------------------------------===//
3889 // SSE2 - Packed Integer Comparison Instructions
3890 //===---------------------------------------------------------------------===//
3892 let Predicates = [HasAVX] in {
3893 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3894 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3895 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3896 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3897 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3898 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3899 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3900 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3901 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3902 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3903 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3904 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3906 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3907 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3908 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3909 (bc_v16i8 (memopv2i64 addr:$src2)))),
3910 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3911 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3912 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3913 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3914 (bc_v8i16 (memopv2i64 addr:$src2)))),
3915 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3916 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3917 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3918 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3919 (bc_v4i32 (memopv2i64 addr:$src2)))),
3920 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3922 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3923 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3924 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3925 (bc_v16i8 (memopv2i64 addr:$src2)))),
3926 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3927 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3928 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3929 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3930 (bc_v8i16 (memopv2i64 addr:$src2)))),
3931 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3932 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3933 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3934 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3935 (bc_v4i32 (memopv2i64 addr:$src2)))),
3936 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3939 let Predicates = [HasAVX2] in {
3940 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3941 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3942 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3943 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3944 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3945 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3946 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3947 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3948 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3949 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3950 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3951 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3953 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3954 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3955 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3956 (bc_v32i8 (memopv4i64 addr:$src2)))),
3957 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3958 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3959 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3960 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3961 (bc_v16i16 (memopv4i64 addr:$src2)))),
3962 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3963 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3964 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3965 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3966 (bc_v8i32 (memopv4i64 addr:$src2)))),
3967 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3969 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3970 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3971 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3972 (bc_v32i8 (memopv4i64 addr:$src2)))),
3973 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3974 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3975 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3976 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3977 (bc_v16i16 (memopv4i64 addr:$src2)))),
3978 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3979 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3980 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3981 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3982 (bc_v8i32 (memopv4i64 addr:$src2)))),
3983 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3986 let Constraints = "$src1 = $dst" in {
3987 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3988 VR128, memopv2i64, i128mem, 1>;
3989 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3990 VR128, memopv2i64, i128mem, 1>;
3991 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3992 VR128, memopv2i64, i128mem, 1>;
3993 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3994 VR128, memopv2i64, i128mem>;
3995 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3996 VR128, memopv2i64, i128mem>;
3997 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3998 VR128, memopv2i64, i128mem>;
3999 } // Constraints = "$src1 = $dst"
4001 let Predicates = [HasSSE2] in {
4002 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
4003 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4004 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
4005 (bc_v16i8 (memopv2i64 addr:$src2)))),
4006 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4007 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4008 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4009 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4010 (bc_v8i16 (memopv2i64 addr:$src2)))),
4011 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4012 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4013 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4014 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4015 (bc_v4i32 (memopv2i64 addr:$src2)))),
4016 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4018 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4019 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4020 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4021 (bc_v16i8 (memopv2i64 addr:$src2)))),
4022 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4023 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4024 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4025 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4026 (bc_v8i16 (memopv2i64 addr:$src2)))),
4027 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4028 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4029 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4030 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4031 (bc_v4i32 (memopv2i64 addr:$src2)))),
4032 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4035 //===---------------------------------------------------------------------===//
4036 // SSE2 - Packed Integer Pack Instructions
4037 //===---------------------------------------------------------------------===//
4039 let Predicates = [HasAVX] in {
4040 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4041 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4042 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4043 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4044 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4045 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4048 let Predicates = [HasAVX2] in {
4049 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4050 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4051 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4052 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4053 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4054 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4057 let Constraints = "$src1 = $dst" in {
4058 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4059 VR128, memopv2i64, i128mem>;
4060 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4061 VR128, memopv2i64, i128mem>;
4062 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4063 VR128, memopv2i64, i128mem>;
4064 } // Constraints = "$src1 = $dst"
4066 //===---------------------------------------------------------------------===//
4067 // SSE2 - Packed Integer Shuffle Instructions
4068 //===---------------------------------------------------------------------===//
4070 let ExeDomain = SSEPackedInt in {
4071 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4073 def ri : Ii8<0x70, MRMSrcReg,
4074 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4075 !strconcat(OpcodeStr,
4076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4077 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4079 def mi : Ii8<0x70, MRMSrcMem,
4080 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4081 !strconcat(OpcodeStr,
4082 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4083 [(set VR128:$dst, (vt (pshuf_frag:$src2
4084 (bc_frag (memopv2i64 addr:$src1)),
4088 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4090 def Yri : Ii8<0x70, MRMSrcReg,
4091 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4092 !strconcat(OpcodeStr,
4093 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4094 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4096 def Ymi : Ii8<0x70, MRMSrcMem,
4097 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4098 !strconcat(OpcodeStr,
4099 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4100 [(set VR256:$dst, (vt (pshuf_frag:$src2
4101 (bc_frag (memopv4i64 addr:$src1)),
4104 } // ExeDomain = SSEPackedInt
4106 let Predicates = [HasAVX] in {
4107 let AddedComplexity = 5 in
4108 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4111 // SSE2 with ImmT == Imm8 and XS prefix.
4112 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4115 // SSE2 with ImmT == Imm8 and XD prefix.
4116 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4119 let AddedComplexity = 5 in
4120 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4121 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4122 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4123 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4124 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4126 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4128 (VPSHUFDmi addr:$src1, imm:$imm)>;
4129 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4131 (VPSHUFDmi addr:$src1, imm:$imm)>;
4132 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4133 (VPSHUFDri VR128:$src1, imm:$imm)>;
4134 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4135 (VPSHUFDri VR128:$src1, imm:$imm)>;
4136 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4137 (VPSHUFHWri VR128:$src, imm:$imm)>;
4138 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4140 (VPSHUFHWmi addr:$src, imm:$imm)>;
4141 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4142 (VPSHUFLWri VR128:$src, imm:$imm)>;
4143 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4145 (VPSHUFLWmi addr:$src, imm:$imm)>;
4148 let Predicates = [HasAVX2] in {
4149 let AddedComplexity = 5 in
4150 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4153 // SSE2 with ImmT == Imm8 and XS prefix.
4154 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4157 // SSE2 with ImmT == Imm8 and XD prefix.
4158 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4162 let Predicates = [HasSSE2] in {
4163 let AddedComplexity = 5 in
4164 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4166 // SSE2 with ImmT == Imm8 and XS prefix.
4167 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4169 // SSE2 with ImmT == Imm8 and XD prefix.
4170 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4172 let AddedComplexity = 5 in
4173 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4174 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4175 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4176 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4177 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4179 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4181 (PSHUFDmi addr:$src1, imm:$imm)>;
4182 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4184 (PSHUFDmi addr:$src1, imm:$imm)>;
4185 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4186 (PSHUFDri VR128:$src1, imm:$imm)>;
4187 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4188 (PSHUFDri VR128:$src1, imm:$imm)>;
4189 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4190 (PSHUFHWri VR128:$src, imm:$imm)>;
4191 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4193 (PSHUFHWmi addr:$src, imm:$imm)>;
4194 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4195 (PSHUFLWri VR128:$src, imm:$imm)>;
4196 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4198 (PSHUFLWmi addr:$src, imm:$imm)>;
4201 //===---------------------------------------------------------------------===//
4202 // SSE2 - Packed Integer Unpack Instructions
4203 //===---------------------------------------------------------------------===//
4205 let ExeDomain = SSEPackedInt in {
4206 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4207 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4208 def rr : PDI<opc, MRMSrcReg,
4209 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4211 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4212 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4213 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4214 def rm : PDI<opc, MRMSrcMem,
4215 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4217 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4218 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4219 [(set VR128:$dst, (OpNode VR128:$src1,
4220 (bc_frag (memopv2i64
4224 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4225 SDNode OpNode, PatFrag bc_frag> {
4226 def Yrr : PDI<opc, MRMSrcReg,
4227 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4228 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4229 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4230 def Yrm : PDI<opc, MRMSrcMem,
4231 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4232 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4233 [(set VR256:$dst, (OpNode VR256:$src1,
4234 (bc_frag (memopv4i64 addr:$src2))))]>;
4237 let Predicates = [HasAVX] in {
4238 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4239 bc_v16i8, 0>, VEX_4V;
4240 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4241 bc_v8i16, 0>, VEX_4V;
4242 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4243 bc_v4i32, 0>, VEX_4V;
4244 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4245 bc_v2i64, 0>, VEX_4V;
4247 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4248 bc_v16i8, 0>, VEX_4V;
4249 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4250 bc_v8i16, 0>, VEX_4V;
4251 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4252 bc_v4i32, 0>, VEX_4V;
4253 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4254 bc_v2i64, 0>, VEX_4V;
4257 let Predicates = [HasAVX2] in {
4258 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4260 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4262 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4264 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4267 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4269 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4271 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4273 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4277 let Constraints = "$src1 = $dst" in {
4278 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4280 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4282 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4284 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4287 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4289 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4291 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4293 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4296 } // ExeDomain = SSEPackedInt
4298 // Patterns for using AVX1 instructions with integer vectors
4299 // Here to give AVX2 priority
4300 let Predicates = [HasAVX] in {
4301 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4302 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4303 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4304 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4305 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4306 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4307 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4308 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4310 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4311 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4312 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4313 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4314 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4315 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4316 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4317 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4320 // Splat v2f64 / v2i64
4321 let AddedComplexity = 10 in {
4322 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4323 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4324 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4325 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4328 //===---------------------------------------------------------------------===//
4329 // SSE2 - Packed Integer Extract and Insert
4330 //===---------------------------------------------------------------------===//
4332 let ExeDomain = SSEPackedInt in {
4333 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4334 def rri : Ii8<0xC4, MRMSrcReg,
4335 (outs VR128:$dst), (ins VR128:$src1,
4336 GR32:$src2, i32i8imm:$src3),
4338 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4339 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4341 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4342 def rmi : Ii8<0xC4, MRMSrcMem,
4343 (outs VR128:$dst), (ins VR128:$src1,
4344 i16mem:$src2, i32i8imm:$src3),
4346 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4347 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4349 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4354 let Predicates = [HasAVX] in
4355 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4356 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4357 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4358 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4359 imm:$src2))]>, TB, OpSize, VEX;
4360 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4361 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4362 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4363 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4367 let Predicates = [HasAVX] in {
4368 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4369 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4370 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4371 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4372 []>, TB, OpSize, VEX_4V;
4375 let Constraints = "$src1 = $dst" in
4376 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4378 } // ExeDomain = SSEPackedInt
4380 //===---------------------------------------------------------------------===//
4381 // SSE2 - Packed Mask Creation
4382 //===---------------------------------------------------------------------===//
4384 let ExeDomain = SSEPackedInt in {
4386 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4387 "pmovmskb\t{$src, $dst|$dst, $src}",
4388 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4389 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4390 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4392 let Predicates = [HasAVX2] in {
4393 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4394 "pmovmskb\t{$src, $dst|$dst, $src}",
4395 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4396 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4397 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4400 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4401 "pmovmskb\t{$src, $dst|$dst, $src}",
4402 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4404 } // ExeDomain = SSEPackedInt
4406 //===---------------------------------------------------------------------===//
4407 // SSE2 - Conditional Store
4408 //===---------------------------------------------------------------------===//
4410 let ExeDomain = SSEPackedInt in {
4413 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4414 (ins VR128:$src, VR128:$mask),
4415 "maskmovdqu\t{$mask, $src|$src, $mask}",
4416 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4418 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4419 (ins VR128:$src, VR128:$mask),
4420 "maskmovdqu\t{$mask, $src|$src, $mask}",
4421 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4424 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4425 "maskmovdqu\t{$mask, $src|$src, $mask}",
4426 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4428 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4429 "maskmovdqu\t{$mask, $src|$src, $mask}",
4430 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4432 } // ExeDomain = SSEPackedInt
4434 //===---------------------------------------------------------------------===//
4435 // SSE2 - Move Doubleword
4436 //===---------------------------------------------------------------------===//
4438 //===---------------------------------------------------------------------===//
4439 // Move Int Doubleword to Packed Double Int
4441 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4442 "movd\t{$src, $dst|$dst, $src}",
4444 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4445 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4446 "movd\t{$src, $dst|$dst, $src}",
4448 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4450 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4451 "mov{d|q}\t{$src, $dst|$dst, $src}",
4453 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4454 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4455 "mov{d|q}\t{$src, $dst|$dst, $src}",
4456 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4458 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4459 "movd\t{$src, $dst|$dst, $src}",
4461 (v4i32 (scalar_to_vector GR32:$src)))]>;
4462 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4463 "movd\t{$src, $dst|$dst, $src}",
4465 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4466 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4467 "mov{d|q}\t{$src, $dst|$dst, $src}",
4469 (v2i64 (scalar_to_vector GR64:$src)))]>;
4470 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4471 "mov{d|q}\t{$src, $dst|$dst, $src}",
4472 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4474 //===---------------------------------------------------------------------===//
4475 // Move Int Doubleword to Single Scalar
4477 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4478 "movd\t{$src, $dst|$dst, $src}",
4479 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4481 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4482 "movd\t{$src, $dst|$dst, $src}",
4483 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4485 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4486 "movd\t{$src, $dst|$dst, $src}",
4487 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4489 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4490 "movd\t{$src, $dst|$dst, $src}",
4491 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4493 //===---------------------------------------------------------------------===//
4494 // Move Packed Doubleword Int to Packed Double Int
4496 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4497 "movd\t{$src, $dst|$dst, $src}",
4498 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4500 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4501 (ins i32mem:$dst, VR128:$src),
4502 "movd\t{$src, $dst|$dst, $src}",
4503 [(store (i32 (vector_extract (v4i32 VR128:$src),
4504 (iPTR 0))), addr:$dst)]>, VEX;
4505 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4506 "movd\t{$src, $dst|$dst, $src}",
4507 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4509 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4510 "movd\t{$src, $dst|$dst, $src}",
4511 [(store (i32 (vector_extract (v4i32 VR128:$src),
4512 (iPTR 0))), addr:$dst)]>;
4514 //===---------------------------------------------------------------------===//
4515 // Move Packed Doubleword Int first element to Doubleword Int
4517 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4518 "mov{d|q}\t{$src, $dst|$dst, $src}",
4519 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4521 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4523 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4524 "mov{d|q}\t{$src, $dst|$dst, $src}",
4525 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4528 //===---------------------------------------------------------------------===//
4529 // Bitcast FR64 <-> GR64
4531 let Predicates = [HasAVX] in
4532 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4533 "vmovq\t{$src, $dst|$dst, $src}",
4534 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4536 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4537 "mov{d|q}\t{$src, $dst|$dst, $src}",
4538 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4539 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4540 "movq\t{$src, $dst|$dst, $src}",
4541 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4543 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4544 "movq\t{$src, $dst|$dst, $src}",
4545 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4546 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4547 "mov{d|q}\t{$src, $dst|$dst, $src}",
4548 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4549 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4550 "movq\t{$src, $dst|$dst, $src}",
4551 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4553 //===---------------------------------------------------------------------===//
4554 // Move Scalar Single to Double Int
4556 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4557 "movd\t{$src, $dst|$dst, $src}",
4558 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4559 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4560 "movd\t{$src, $dst|$dst, $src}",
4561 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4562 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4563 "movd\t{$src, $dst|$dst, $src}",
4564 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4565 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4566 "movd\t{$src, $dst|$dst, $src}",
4567 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4569 //===---------------------------------------------------------------------===//
4570 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4572 let AddedComplexity = 15 in {
4573 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4574 "movd\t{$src, $dst|$dst, $src}",
4575 [(set VR128:$dst, (v4i32 (X86vzmovl
4576 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4578 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4579 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4580 [(set VR128:$dst, (v2i64 (X86vzmovl
4581 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4584 let AddedComplexity = 15 in {
4585 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4586 "movd\t{$src, $dst|$dst, $src}",
4587 [(set VR128:$dst, (v4i32 (X86vzmovl
4588 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4589 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4590 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4591 [(set VR128:$dst, (v2i64 (X86vzmovl
4592 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4595 let AddedComplexity = 20 in {
4596 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4597 "movd\t{$src, $dst|$dst, $src}",
4599 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4600 (loadi32 addr:$src))))))]>,
4602 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4603 "movd\t{$src, $dst|$dst, $src}",
4605 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4606 (loadi32 addr:$src))))))]>;
4609 let Predicates = [HasAVX] in {
4610 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4611 let AddedComplexity = 20 in {
4612 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4613 (VMOVZDI2PDIrm addr:$src)>;
4614 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4615 (VMOVZDI2PDIrm addr:$src)>;
4616 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4617 (VMOVZDI2PDIrm addr:$src)>;
4619 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4620 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4621 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4622 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4623 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4624 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4625 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4628 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4629 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4630 (MOVZDI2PDIrm addr:$src)>;
4631 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4632 (MOVZDI2PDIrm addr:$src)>;
4633 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4634 (MOVZDI2PDIrm addr:$src)>;
4637 // These are the correct encodings of the instructions so that we know how to
4638 // read correct assembly, even though we continue to emit the wrong ones for
4639 // compatibility with Darwin's buggy assembler.
4640 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4641 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4642 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4643 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4644 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4645 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4646 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4647 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4648 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4649 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4650 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4651 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4653 //===---------------------------------------------------------------------===//
4654 // SSE2 - Move Quadword
4655 //===---------------------------------------------------------------------===//
4657 //===---------------------------------------------------------------------===//
4658 // Move Quadword Int to Packed Quadword Int
4660 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4661 "vmovq\t{$src, $dst|$dst, $src}",
4663 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4664 VEX, Requires<[HasAVX]>;
4665 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4666 "movq\t{$src, $dst|$dst, $src}",
4668 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4669 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4671 //===---------------------------------------------------------------------===//
4672 // Move Packed Quadword Int to Quadword Int
4674 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4675 "movq\t{$src, $dst|$dst, $src}",
4676 [(store (i64 (vector_extract (v2i64 VR128:$src),
4677 (iPTR 0))), addr:$dst)]>, VEX;
4678 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4679 "movq\t{$src, $dst|$dst, $src}",
4680 [(store (i64 (vector_extract (v2i64 VR128:$src),
4681 (iPTR 0))), addr:$dst)]>;
4683 //===---------------------------------------------------------------------===//
4684 // Store / copy lower 64-bits of a XMM register.
4686 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4687 "movq\t{$src, $dst|$dst, $src}",
4688 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4689 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4690 "movq\t{$src, $dst|$dst, $src}",
4691 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4693 let AddedComplexity = 20 in
4694 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4695 "vmovq\t{$src, $dst|$dst, $src}",
4697 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4698 (loadi64 addr:$src))))))]>,
4699 XS, VEX, Requires<[HasAVX]>;
4701 let AddedComplexity = 20 in
4702 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4703 "movq\t{$src, $dst|$dst, $src}",
4705 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4706 (loadi64 addr:$src))))))]>,
4707 XS, Requires<[HasSSE2]>;
4709 let Predicates = [HasAVX], AddedComplexity = 20 in {
4710 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4711 (VMOVZQI2PQIrm addr:$src)>;
4712 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4713 (VMOVZQI2PQIrm addr:$src)>;
4714 def : Pat<(v2i64 (X86vzload addr:$src)),
4715 (VMOVZQI2PQIrm addr:$src)>;
4718 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4719 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4720 (MOVZQI2PQIrm addr:$src)>;
4721 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4722 (MOVZQI2PQIrm addr:$src)>;
4723 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4726 let Predicates = [HasAVX] in {
4727 def : Pat<(v4i64 (X86vzload addr:$src)),
4728 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4731 //===---------------------------------------------------------------------===//
4732 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4733 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4735 let AddedComplexity = 15 in
4736 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4737 "vmovq\t{$src, $dst|$dst, $src}",
4738 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4739 XS, VEX, Requires<[HasAVX]>;
4740 let AddedComplexity = 15 in
4741 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4742 "movq\t{$src, $dst|$dst, $src}",
4743 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4744 XS, Requires<[HasSSE2]>;
4746 let AddedComplexity = 20 in
4747 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4748 "vmovq\t{$src, $dst|$dst, $src}",
4749 [(set VR128:$dst, (v2i64 (X86vzmovl
4750 (loadv2i64 addr:$src))))]>,
4751 XS, VEX, Requires<[HasAVX]>;
4752 let AddedComplexity = 20 in {
4753 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4754 "movq\t{$src, $dst|$dst, $src}",
4755 [(set VR128:$dst, (v2i64 (X86vzmovl
4756 (loadv2i64 addr:$src))))]>,
4757 XS, Requires<[HasSSE2]>;
4760 let AddedComplexity = 20 in {
4761 let Predicates = [HasAVX] in {
4762 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4763 (VMOVZPQILo2PQIrm addr:$src)>;
4764 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4765 (VMOVZPQILo2PQIrr VR128:$src)>;
4767 let Predicates = [HasSSE2] in {
4768 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4769 (MOVZPQILo2PQIrm addr:$src)>;
4770 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4771 (MOVZPQILo2PQIrr VR128:$src)>;
4775 // Instructions to match in the assembler
4776 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4777 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4778 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4779 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4780 // Recognize "movd" with GR64 destination, but encode as a "movq"
4781 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4782 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4784 // Instructions for the disassembler
4785 // xr = XMM register
4788 let Predicates = [HasAVX] in
4789 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4790 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4791 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4792 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4794 //===---------------------------------------------------------------------===//
4795 // SSE3 - Conversion Instructions
4796 //===---------------------------------------------------------------------===//
4798 // Convert Packed Double FP to Packed DW Integers
4799 let Predicates = [HasAVX] in {
4800 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4801 // register, but the same isn't true when using memory operands instead.
4802 // Provide other assembly rr and rm forms to address this explicitly.
4803 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4804 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4805 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4806 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4809 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4810 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4811 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4812 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4815 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4816 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4817 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4818 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4821 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4822 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4823 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4824 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4826 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4827 (VCVTPD2DQYrr VR256:$src)>;
4828 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4829 (VCVTPD2DQYrm addr:$src)>;
4831 // Convert Packed DW Integers to Packed Double FP
4832 let Predicates = [HasAVX] in {
4833 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4834 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4835 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4836 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4837 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4838 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4839 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4840 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4843 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4844 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4845 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4846 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4848 // AVX 256-bit register conversion intrinsics
4849 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4850 (VCVTDQ2PDYrr VR128:$src)>;
4851 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4852 (VCVTDQ2PDYrm addr:$src)>;
4854 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4855 (VCVTPD2DQYrr VR256:$src)>;
4856 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4857 (VCVTPD2DQYrm addr:$src)>;
4859 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4860 (VCVTDQ2PDYrr VR128:$src)>;
4861 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4862 (VCVTDQ2PDYrm addr:$src)>;
4864 //===---------------------------------------------------------------------===//
4865 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4866 //===---------------------------------------------------------------------===//
4867 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4868 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4869 X86MemOperand x86memop> {
4870 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4871 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4872 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4873 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4874 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4875 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4878 let Predicates = [HasAVX] in {
4879 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4880 v4f32, VR128, memopv4f32, f128mem>, VEX;
4881 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4882 v4f32, VR128, memopv4f32, f128mem>, VEX;
4883 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4884 v8f32, VR256, memopv8f32, f256mem>, VEX;
4885 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4886 v8f32, VR256, memopv8f32, f256mem>, VEX;
4888 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4889 memopv4f32, f128mem>;
4890 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4891 memopv4f32, f128mem>;
4893 let Predicates = [HasAVX] in {
4894 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4895 (VMOVSHDUPrr VR128:$src)>;
4896 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4897 (VMOVSHDUPrm addr:$src)>;
4898 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4899 (VMOVSLDUPrr VR128:$src)>;
4900 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4901 (VMOVSLDUPrm addr:$src)>;
4902 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4903 (VMOVSHDUPYrr VR256:$src)>;
4904 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4905 (VMOVSHDUPYrm addr:$src)>;
4906 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4907 (VMOVSLDUPYrr VR256:$src)>;
4908 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4909 (VMOVSLDUPYrm addr:$src)>;
4912 let Predicates = [HasSSE3] in {
4913 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4914 (MOVSHDUPrr VR128:$src)>;
4915 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4916 (MOVSHDUPrm addr:$src)>;
4917 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4918 (MOVSLDUPrr VR128:$src)>;
4919 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4920 (MOVSLDUPrm addr:$src)>;
4923 //===---------------------------------------------------------------------===//
4924 // SSE3 - Replicate Double FP - MOVDDUP
4925 //===---------------------------------------------------------------------===//
4927 multiclass sse3_replicate_dfp<string OpcodeStr> {
4928 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4929 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4930 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4931 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4932 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4934 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4938 // FIXME: Merge with above classe when there're patterns for the ymm version
4939 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4940 let Predicates = [HasAVX] in {
4941 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4942 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4944 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4945 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4950 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4951 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4952 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4954 let Predicates = [HasAVX] in {
4955 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4957 (VMOVDDUPrm addr:$src)>;
4958 let AddedComplexity = 5 in {
4959 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4960 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4961 (VMOVDDUPrm addr:$src)>;
4962 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4963 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4964 (VMOVDDUPrm addr:$src)>;
4966 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4967 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4968 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4969 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4970 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4971 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4972 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4973 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4974 def : Pat<(X86Movddup (bc_v2f64
4975 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4976 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4979 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4980 (VMOVDDUPYrm addr:$src)>;
4981 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4982 (VMOVDDUPYrm addr:$src)>;
4983 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4984 (VMOVDDUPYrm addr:$src)>;
4985 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4986 (VMOVDDUPYrm addr:$src)>;
4987 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4988 (VMOVDDUPYrr VR256:$src)>;
4989 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4990 (VMOVDDUPYrr VR256:$src)>;
4993 let Predicates = [HasSSE3] in {
4994 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4996 (MOVDDUPrm addr:$src)>;
4997 let AddedComplexity = 5 in {
4998 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4999 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
5000 (MOVDDUPrm addr:$src)>;
5001 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
5002 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5003 (MOVDDUPrm addr:$src)>;
5005 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5006 (MOVDDUPrm addr:$src)>;
5007 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5008 (MOVDDUPrm addr:$src)>;
5009 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5010 (MOVDDUPrm addr:$src)>;
5011 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5012 (MOVDDUPrm addr:$src)>;
5013 def : Pat<(X86Movddup (bc_v2f64
5014 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5015 (MOVDDUPrm addr:$src)>;
5018 //===---------------------------------------------------------------------===//
5019 // SSE3 - Move Unaligned Integer
5020 //===---------------------------------------------------------------------===//
5022 let Predicates = [HasAVX] in {
5023 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5024 "vlddqu\t{$src, $dst|$dst, $src}",
5025 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5026 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5027 "vlddqu\t{$src, $dst|$dst, $src}",
5028 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5030 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5031 "lddqu\t{$src, $dst|$dst, $src}",
5032 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5034 //===---------------------------------------------------------------------===//
5035 // SSE3 - Arithmetic
5036 //===---------------------------------------------------------------------===//
5038 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5039 X86MemOperand x86memop, bit Is2Addr = 1> {
5040 def rr : I<0xD0, MRMSrcReg,
5041 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5043 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5044 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5045 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5046 def rm : I<0xD0, MRMSrcMem,
5047 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5049 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5050 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5051 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5054 let Predicates = [HasAVX] in {
5055 let ExeDomain = SSEPackedSingle in {
5056 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5057 f128mem, 0>, TB, XD, VEX_4V;
5058 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5059 f256mem, 0>, TB, XD, VEX_4V;
5061 let ExeDomain = SSEPackedDouble in {
5062 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5063 f128mem, 0>, TB, OpSize, VEX_4V;
5064 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5065 f256mem, 0>, TB, OpSize, VEX_4V;
5068 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5069 let ExeDomain = SSEPackedSingle in
5070 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5072 let ExeDomain = SSEPackedDouble in
5073 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5074 f128mem>, TB, OpSize;
5077 //===---------------------------------------------------------------------===//
5078 // SSE3 Instructions
5079 //===---------------------------------------------------------------------===//
5082 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5083 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5084 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5086 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5087 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5088 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5090 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5092 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5093 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5094 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5096 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5097 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5098 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5100 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5101 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5102 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5104 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5106 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5107 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5108 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5111 let Predicates = [HasAVX] in {
5112 let ExeDomain = SSEPackedSingle in {
5113 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5114 X86fhadd, 0>, VEX_4V;
5115 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5116 X86fhsub, 0>, VEX_4V;
5117 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5118 X86fhadd, 0>, VEX_4V;
5119 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5120 X86fhsub, 0>, VEX_4V;
5122 let ExeDomain = SSEPackedDouble in {
5123 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5124 X86fhadd, 0>, VEX_4V;
5125 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5126 X86fhsub, 0>, VEX_4V;
5127 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5128 X86fhadd, 0>, VEX_4V;
5129 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5130 X86fhsub, 0>, VEX_4V;
5134 let Constraints = "$src1 = $dst" in {
5135 let ExeDomain = SSEPackedSingle in {
5136 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5137 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5139 let ExeDomain = SSEPackedDouble in {
5140 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5141 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5145 //===---------------------------------------------------------------------===//
5146 // SSSE3 - Packed Absolute Instructions
5147 //===---------------------------------------------------------------------===//
5150 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5151 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5152 Intrinsic IntId128> {
5153 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5155 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5156 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5159 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5161 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5164 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5167 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5168 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5169 Intrinsic IntId256> {
5170 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5172 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5173 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5176 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5178 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5181 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5184 let Predicates = [HasAVX] in {
5185 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5186 int_x86_ssse3_pabs_b_128>, VEX;
5187 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5188 int_x86_ssse3_pabs_w_128>, VEX;
5189 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5190 int_x86_ssse3_pabs_d_128>, VEX;
5193 let Predicates = [HasAVX2] in {
5194 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5195 int_x86_avx2_pabs_b>, VEX;
5196 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5197 int_x86_avx2_pabs_w>, VEX;
5198 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5199 int_x86_avx2_pabs_d>, VEX;
5202 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5203 int_x86_ssse3_pabs_b_128>;
5204 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5205 int_x86_ssse3_pabs_w_128>;
5206 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5207 int_x86_ssse3_pabs_d_128>;
5209 //===---------------------------------------------------------------------===//
5210 // SSSE3 - Packed Binary Operator Instructions
5211 //===---------------------------------------------------------------------===//
5213 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5214 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5215 Intrinsic IntId128, bit Is2Addr = 1> {
5216 let isCommutable = 1 in
5217 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5218 (ins VR128:$src1, VR128:$src2),
5220 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5221 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5222 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5224 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5225 (ins VR128:$src1, i128mem:$src2),
5227 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5228 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5230 (IntId128 VR128:$src1,
5231 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5234 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5235 Intrinsic IntId256> {
5236 let isCommutable = 1 in
5237 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5238 (ins VR256:$src1, VR256:$src2),
5239 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5240 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5242 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5243 (ins VR256:$src1, i256mem:$src2),
5244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5246 (IntId256 VR256:$src1,
5247 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5250 let ImmT = NoImm, Predicates = [HasAVX] in {
5251 let isCommutable = 0 in {
5252 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5253 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5254 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5255 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5256 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5257 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5258 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5259 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5260 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5261 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5262 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5263 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5264 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5265 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5266 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5267 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5268 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5269 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5270 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5271 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5272 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5273 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5275 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5276 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5279 let ImmT = NoImm, Predicates = [HasAVX2] in {
5280 let isCommutable = 0 in {
5281 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5282 int_x86_avx2_phadd_w>, VEX_4V;
5283 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5284 int_x86_avx2_phadd_d>, VEX_4V;
5285 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5286 int_x86_avx2_phadd_sw>, VEX_4V;
5287 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5288 int_x86_avx2_phsub_w>, VEX_4V;
5289 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5290 int_x86_avx2_phsub_d>, VEX_4V;
5291 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5292 int_x86_avx2_phsub_sw>, VEX_4V;
5293 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5294 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5295 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5296 int_x86_avx2_pshuf_b>, VEX_4V;
5297 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5298 int_x86_avx2_psign_b>, VEX_4V;
5299 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5300 int_x86_avx2_psign_w>, VEX_4V;
5301 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5302 int_x86_avx2_psign_d>, VEX_4V;
5304 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5305 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5308 // None of these have i8 immediate fields.
5309 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5310 let isCommutable = 0 in {
5311 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5312 int_x86_ssse3_phadd_w_128>;
5313 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5314 int_x86_ssse3_phadd_d_128>;
5315 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5316 int_x86_ssse3_phadd_sw_128>;
5317 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5318 int_x86_ssse3_phsub_w_128>;
5319 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5320 int_x86_ssse3_phsub_d_128>;
5321 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5322 int_x86_ssse3_phsub_sw_128>;
5323 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5324 int_x86_ssse3_pmadd_ub_sw_128>;
5325 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5326 int_x86_ssse3_pshuf_b_128>;
5327 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5328 int_x86_ssse3_psign_b_128>;
5329 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5330 int_x86_ssse3_psign_w_128>;
5331 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5332 int_x86_ssse3_psign_d_128>;
5334 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5335 int_x86_ssse3_pmul_hr_sw_128>;
5338 let Predicates = [HasAVX] in {
5339 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5340 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5341 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5342 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5344 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5345 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5346 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5347 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5348 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5349 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5351 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5352 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5353 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5354 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5355 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5356 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5357 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5358 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5361 let Predicates = [HasAVX2] in {
5362 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5363 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5364 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5365 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5366 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5367 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5369 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5370 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5371 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5372 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5373 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5374 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5375 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5376 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5379 let Predicates = [HasSSSE3] in {
5380 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5381 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5382 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5383 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5385 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5386 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5387 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5388 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5389 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5390 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5392 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5393 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5394 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5395 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5396 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5397 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5398 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5399 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5402 //===---------------------------------------------------------------------===//
5403 // SSSE3 - Packed Align Instruction Patterns
5404 //===---------------------------------------------------------------------===//
5406 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5407 let neverHasSideEffects = 1 in {
5408 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5409 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5411 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5413 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5416 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5417 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5419 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5421 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5426 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5427 let neverHasSideEffects = 1 in {
5428 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5429 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5431 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5434 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5435 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5437 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5442 let Predicates = [HasAVX] in
5443 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5444 let Predicates = [HasAVX2] in
5445 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5446 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5447 defm PALIGN : ssse3_palign<"palignr">;
5449 let Predicates = [HasAVX] in {
5450 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5451 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5452 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5453 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5454 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5455 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5456 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5457 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5460 let Predicates = [HasSSSE3] in {
5461 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5462 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5463 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5464 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5465 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5466 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5467 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5468 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5471 //===---------------------------------------------------------------------===//
5472 // SSSE3 - Thread synchronization
5473 //===---------------------------------------------------------------------===//
5475 let usesCustomInserter = 1 in {
5476 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5477 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5478 Requires<[HasSSE3orAVX]>;
5479 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5480 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5481 Requires<[HasSSE3orAVX]>;
5484 let Uses = [EAX, ECX, EDX] in
5485 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5486 Requires<[HasSSE3orAVX]>;
5487 let Uses = [ECX, EAX] in
5488 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5489 Requires<[HasSSE3orAVX]>;
5491 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5492 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5494 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5495 Requires<[In32BitMode]>;
5496 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5497 Requires<[In64BitMode]>;
5499 //===----------------------------------------------------------------------===//
5500 // SSE4.1 - Packed Move with Sign/Zero Extend
5501 //===----------------------------------------------------------------------===//
5503 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5504 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5505 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5506 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5508 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5509 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5511 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5515 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5517 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5518 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5519 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5521 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5522 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5523 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5526 let Predicates = [HasAVX] in {
5527 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5529 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5531 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5533 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5535 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5537 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5541 let Predicates = [HasAVX2] in {
5542 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5543 int_x86_avx2_pmovsxbw>, VEX;
5544 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5545 int_x86_avx2_pmovsxwd>, VEX;
5546 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5547 int_x86_avx2_pmovsxdq>, VEX;
5548 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5549 int_x86_avx2_pmovzxbw>, VEX;
5550 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5551 int_x86_avx2_pmovzxwd>, VEX;
5552 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5553 int_x86_avx2_pmovzxdq>, VEX;
5556 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5557 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5558 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5559 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5560 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5561 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5563 let Predicates = [HasAVX] in {
5564 // Common patterns involving scalar load.
5565 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5566 (VPMOVSXBWrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5568 (VPMOVSXBWrm addr:$src)>;
5570 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5571 (VPMOVSXWDrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5573 (VPMOVSXWDrm addr:$src)>;
5575 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5576 (VPMOVSXDQrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5578 (VPMOVSXDQrm addr:$src)>;
5580 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5581 (VPMOVZXBWrm addr:$src)>;
5582 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5583 (VPMOVZXBWrm addr:$src)>;
5585 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5586 (VPMOVZXWDrm addr:$src)>;
5587 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5588 (VPMOVZXWDrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5591 (VPMOVZXDQrm addr:$src)>;
5592 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5593 (VPMOVZXDQrm addr:$src)>;
5596 let Predicates = [HasSSE41] in {
5597 // Common patterns involving scalar load.
5598 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5599 (PMOVSXBWrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5601 (PMOVSXBWrm addr:$src)>;
5603 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5604 (PMOVSXWDrm addr:$src)>;
5605 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5606 (PMOVSXWDrm addr:$src)>;
5608 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5609 (PMOVSXDQrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5611 (PMOVSXDQrm addr:$src)>;
5613 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5614 (PMOVZXBWrm addr:$src)>;
5615 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5616 (PMOVZXBWrm addr:$src)>;
5618 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5619 (PMOVZXWDrm addr:$src)>;
5620 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5621 (PMOVZXWDrm addr:$src)>;
5623 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5624 (PMOVZXDQrm addr:$src)>;
5625 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5626 (PMOVZXDQrm addr:$src)>;
5630 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5631 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5632 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5633 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5635 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5636 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5638 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5642 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5644 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5645 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5646 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5648 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5649 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5651 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5655 let Predicates = [HasAVX] in {
5656 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5658 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5660 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5662 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5666 let Predicates = [HasAVX2] in {
5667 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5668 int_x86_avx2_pmovsxbd>, VEX;
5669 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5670 int_x86_avx2_pmovsxwq>, VEX;
5671 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5672 int_x86_avx2_pmovzxbd>, VEX;
5673 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5674 int_x86_avx2_pmovzxwq>, VEX;
5677 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5678 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5679 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5680 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5682 let Predicates = [HasAVX] in {
5683 // Common patterns involving scalar load
5684 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5685 (VPMOVSXBDrm addr:$src)>;
5686 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5687 (VPMOVSXWQrm addr:$src)>;
5689 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5690 (VPMOVZXBDrm addr:$src)>;
5691 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5692 (VPMOVZXWQrm addr:$src)>;
5695 let Predicates = [HasSSE41] in {
5696 // Common patterns involving scalar load
5697 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5698 (PMOVSXBDrm addr:$src)>;
5699 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5700 (PMOVSXWQrm addr:$src)>;
5702 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5703 (PMOVZXBDrm addr:$src)>;
5704 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5705 (PMOVZXWQrm addr:$src)>;
5708 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5709 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5710 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5711 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5713 // Expecting a i16 load any extended to i32 value.
5714 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5715 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5716 [(set VR128:$dst, (IntId (bitconvert
5717 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5721 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5723 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5724 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5725 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5727 // Expecting a i16 load any extended to i32 value.
5728 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5729 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5730 [(set VR256:$dst, (IntId (bitconvert
5731 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5735 let Predicates = [HasAVX] in {
5736 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5738 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5741 let Predicates = [HasAVX2] in {
5742 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5743 int_x86_avx2_pmovsxbq>, VEX;
5744 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5745 int_x86_avx2_pmovzxbq>, VEX;
5747 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5748 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5750 let Predicates = [HasAVX] in {
5751 // Common patterns involving scalar load
5752 def : Pat<(int_x86_sse41_pmovsxbq
5753 (bitconvert (v4i32 (X86vzmovl
5754 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5755 (VPMOVSXBQrm addr:$src)>;
5757 def : Pat<(int_x86_sse41_pmovzxbq
5758 (bitconvert (v4i32 (X86vzmovl
5759 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5760 (VPMOVZXBQrm addr:$src)>;
5763 let Predicates = [HasSSE41] in {
5764 // Common patterns involving scalar load
5765 def : Pat<(int_x86_sse41_pmovsxbq
5766 (bitconvert (v4i32 (X86vzmovl
5767 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5768 (PMOVSXBQrm addr:$src)>;
5770 def : Pat<(int_x86_sse41_pmovzxbq
5771 (bitconvert (v4i32 (X86vzmovl
5772 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5773 (PMOVZXBQrm addr:$src)>;
5776 //===----------------------------------------------------------------------===//
5777 // SSE4.1 - Extract Instructions
5778 //===----------------------------------------------------------------------===//
5780 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5781 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5782 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5783 (ins VR128:$src1, i32i8imm:$src2),
5784 !strconcat(OpcodeStr,
5785 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5786 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5788 let neverHasSideEffects = 1, mayStore = 1 in
5789 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5790 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5791 !strconcat(OpcodeStr,
5792 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5795 // There's an AssertZext in the way of writing the store pattern
5796 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5799 let Predicates = [HasAVX] in {
5800 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5801 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5802 (ins VR128:$src1, i32i8imm:$src2),
5803 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5806 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5809 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5810 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5811 let neverHasSideEffects = 1, mayStore = 1 in
5812 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5813 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5814 !strconcat(OpcodeStr,
5815 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5818 // There's an AssertZext in the way of writing the store pattern
5819 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5822 let Predicates = [HasAVX] in
5823 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5825 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5828 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5829 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5830 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5831 (ins VR128:$src1, i32i8imm:$src2),
5832 !strconcat(OpcodeStr,
5833 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5835 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5836 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5837 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5838 !strconcat(OpcodeStr,
5839 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5840 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5841 addr:$dst)]>, OpSize;
5844 let Predicates = [HasAVX] in
5845 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5847 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5849 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5850 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5851 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5852 (ins VR128:$src1, i32i8imm:$src2),
5853 !strconcat(OpcodeStr,
5854 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5856 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5857 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5858 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5859 !strconcat(OpcodeStr,
5860 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5861 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5862 addr:$dst)]>, OpSize, REX_W;
5865 let Predicates = [HasAVX] in
5866 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5868 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5870 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5872 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5873 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5874 (ins VR128:$src1, i32i8imm:$src2),
5875 !strconcat(OpcodeStr,
5876 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5878 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5880 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5881 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5882 !strconcat(OpcodeStr,
5883 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5884 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5885 addr:$dst)]>, OpSize;
5888 let ExeDomain = SSEPackedSingle in {
5889 let Predicates = [HasAVX] in {
5890 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5891 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5892 (ins VR128:$src1, i32i8imm:$src2),
5893 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5896 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5899 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5900 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5903 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5905 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5908 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5909 Requires<[HasSSE41]>;
5911 //===----------------------------------------------------------------------===//
5912 // SSE4.1 - Insert Instructions
5913 //===----------------------------------------------------------------------===//
5915 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5916 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5917 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5919 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5921 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5923 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5924 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5925 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5927 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5929 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5931 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5932 imm:$src3))]>, OpSize;
5935 let Predicates = [HasAVX] in
5936 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5937 let Constraints = "$src1 = $dst" in
5938 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5940 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5941 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5942 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5944 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5946 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5948 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5950 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5951 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5953 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5955 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5957 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5958 imm:$src3)))]>, OpSize;
5961 let Predicates = [HasAVX] in
5962 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5963 let Constraints = "$src1 = $dst" in
5964 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5966 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5967 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5968 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5970 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5972 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5974 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5976 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5977 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5979 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5981 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5983 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5984 imm:$src3)))]>, OpSize;
5987 let Predicates = [HasAVX] in
5988 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5989 let Constraints = "$src1 = $dst" in
5990 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5992 // insertps has a few different modes, there's the first two here below which
5993 // are optimized inserts that won't zero arbitrary elements in the destination
5994 // vector. The next one matches the intrinsic and could zero arbitrary elements
5995 // in the target vector.
5996 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5997 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5998 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6000 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6002 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6004 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6006 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6007 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6009 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6011 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6013 (X86insrtps VR128:$src1,
6014 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6015 imm:$src3))]>, OpSize;
6018 let ExeDomain = SSEPackedSingle in {
6019 let Predicates = [HasAVX] in
6020 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6021 let Constraints = "$src1 = $dst" in
6022 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6025 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6026 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6028 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6029 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6030 Requires<[HasSSE41]>;
6032 //===----------------------------------------------------------------------===//
6033 // SSE4.1 - Round Instructions
6034 //===----------------------------------------------------------------------===//
6036 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6037 X86MemOperand x86memop, RegisterClass RC,
6038 PatFrag mem_frag32, PatFrag mem_frag64,
6039 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6040 let ExeDomain = SSEPackedSingle in {
6041 // Intrinsic operation, reg.
6042 // Vector intrinsic operation, reg
6043 def PSr : SS4AIi8<opcps, MRMSrcReg,
6044 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6045 !strconcat(OpcodeStr,
6046 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6047 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6050 // Vector intrinsic operation, mem
6051 def PSm : SS4AIi8<opcps, MRMSrcMem,
6052 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6053 !strconcat(OpcodeStr,
6054 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6056 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6058 } // ExeDomain = SSEPackedSingle
6060 let ExeDomain = SSEPackedDouble in {
6061 // Vector intrinsic operation, reg
6062 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6063 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6064 !strconcat(OpcodeStr,
6065 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6066 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6069 // Vector intrinsic operation, mem
6070 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6071 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6072 !strconcat(OpcodeStr,
6073 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6075 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6077 } // ExeDomain = SSEPackedDouble
6080 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6083 Intrinsic F64Int, bit Is2Addr = 1> {
6084 let ExeDomain = GenericDomain in {
6086 def SSr : SS4AIi8<opcss, MRMSrcReg,
6087 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6089 !strconcat(OpcodeStr,
6090 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6091 !strconcat(OpcodeStr,
6092 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6095 // Intrinsic operation, reg.
6096 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6097 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6099 !strconcat(OpcodeStr,
6100 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6101 !strconcat(OpcodeStr,
6102 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6103 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6106 // Intrinsic operation, mem.
6107 def SSm : SS4AIi8<opcss, MRMSrcMem,
6108 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6110 !strconcat(OpcodeStr,
6111 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6112 !strconcat(OpcodeStr,
6113 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6115 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6119 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6120 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6122 !strconcat(OpcodeStr,
6123 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6124 !strconcat(OpcodeStr,
6125 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6128 // Intrinsic operation, reg.
6129 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6130 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6132 !strconcat(OpcodeStr,
6133 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6134 !strconcat(OpcodeStr,
6135 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6136 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6139 // Intrinsic operation, mem.
6140 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6141 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6143 !strconcat(OpcodeStr,
6144 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6145 !strconcat(OpcodeStr,
6146 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6148 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6150 } // ExeDomain = GenericDomain
6153 // FP round - roundss, roundps, roundsd, roundpd
6154 let Predicates = [HasAVX] in {
6156 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6157 memopv4f32, memopv2f64,
6158 int_x86_sse41_round_ps,
6159 int_x86_sse41_round_pd>, VEX;
6160 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6161 memopv8f32, memopv4f64,
6162 int_x86_avx_round_ps_256,
6163 int_x86_avx_round_pd_256>, VEX;
6164 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6165 int_x86_sse41_round_ss,
6166 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6168 def : Pat<(ffloor FR32:$src),
6169 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6170 def : Pat<(f64 (ffloor FR64:$src)),
6171 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6172 def : Pat<(f32 (fnearbyint FR32:$src)),
6173 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6174 def : Pat<(f64 (fnearbyint FR64:$src)),
6175 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6176 def : Pat<(f32 (fceil FR32:$src)),
6177 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6178 def : Pat<(f64 (fceil FR64:$src)),
6179 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6180 def : Pat<(f32 (frint FR32:$src)),
6181 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6182 def : Pat<(f64 (frint FR64:$src)),
6183 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6184 def : Pat<(f32 (ftrunc FR32:$src)),
6185 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6186 def : Pat<(f64 (ftrunc FR64:$src)),
6187 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6190 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6191 memopv4f32, memopv2f64,
6192 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6193 let Constraints = "$src1 = $dst" in
6194 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6195 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6197 def : Pat<(ffloor FR32:$src),
6198 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6199 def : Pat<(f64 (ffloor FR64:$src)),
6200 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6201 def : Pat<(f32 (fnearbyint FR32:$src)),
6202 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6203 def : Pat<(f64 (fnearbyint FR64:$src)),
6204 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6205 def : Pat<(f32 (fceil FR32:$src)),
6206 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6207 def : Pat<(f64 (fceil FR64:$src)),
6208 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6209 def : Pat<(f32 (frint FR32:$src)),
6210 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6211 def : Pat<(f64 (frint FR64:$src)),
6212 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6213 def : Pat<(f32 (ftrunc FR32:$src)),
6214 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6215 def : Pat<(f64 (ftrunc FR64:$src)),
6216 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6218 //===----------------------------------------------------------------------===//
6219 // SSE4.1 - Packed Bit Test
6220 //===----------------------------------------------------------------------===//
6222 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6223 // the intel intrinsic that corresponds to this.
6224 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6225 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6226 "vptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6229 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6230 "vptest\t{$src2, $src1|$src1, $src2}",
6231 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6234 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6235 "vptest\t{$src2, $src1|$src1, $src2}",
6236 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6238 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6239 "vptest\t{$src2, $src1|$src1, $src2}",
6240 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6244 let Defs = [EFLAGS] in {
6245 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6246 "ptest\t{$src2, $src1|$src1, $src2}",
6247 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6249 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6250 "ptest\t{$src2, $src1|$src1, $src2}",
6251 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6255 // The bit test instructions below are AVX only
6256 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6257 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6258 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6259 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6260 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6261 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6262 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6263 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6267 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6268 let ExeDomain = SSEPackedSingle in {
6269 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6270 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6272 let ExeDomain = SSEPackedDouble in {
6273 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6274 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6278 //===----------------------------------------------------------------------===//
6279 // SSE4.1 - Misc Instructions
6280 //===----------------------------------------------------------------------===//
6282 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6283 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6284 "popcnt{w}\t{$src, $dst|$dst, $src}",
6285 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6287 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6288 "popcnt{w}\t{$src, $dst|$dst, $src}",
6289 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6290 (implicit EFLAGS)]>, OpSize, XS;
6292 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6293 "popcnt{l}\t{$src, $dst|$dst, $src}",
6294 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6296 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6297 "popcnt{l}\t{$src, $dst|$dst, $src}",
6298 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6299 (implicit EFLAGS)]>, XS;
6301 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6302 "popcnt{q}\t{$src, $dst|$dst, $src}",
6303 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6305 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6306 "popcnt{q}\t{$src, $dst|$dst, $src}",
6307 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6308 (implicit EFLAGS)]>, XS;
6313 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6314 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6315 Intrinsic IntId128> {
6316 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6318 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6319 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6320 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6322 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6325 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6328 let Predicates = [HasAVX] in
6329 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6330 int_x86_sse41_phminposuw>, VEX;
6331 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6332 int_x86_sse41_phminposuw>;
6334 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6335 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6336 Intrinsic IntId128, bit Is2Addr = 1> {
6337 let isCommutable = 1 in
6338 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6339 (ins VR128:$src1, VR128:$src2),
6341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6343 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6344 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6345 (ins VR128:$src1, i128mem:$src2),
6347 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6350 (IntId128 VR128:$src1,
6351 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6354 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6355 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6356 Intrinsic IntId256> {
6357 let isCommutable = 1 in
6358 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6359 (ins VR256:$src1, VR256:$src2),
6360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6361 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6362 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6363 (ins VR256:$src1, i256mem:$src2),
6364 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6366 (IntId256 VR256:$src1,
6367 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6370 let Predicates = [HasAVX] in {
6371 let isCommutable = 0 in
6372 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6374 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6376 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6378 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6380 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6382 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6384 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6386 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6388 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6390 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6392 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6395 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6396 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6397 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6398 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6401 let Predicates = [HasAVX2] in {
6402 let isCommutable = 0 in
6403 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6404 int_x86_avx2_packusdw>, VEX_4V;
6405 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6406 int_x86_avx2_pcmpeq_q>, VEX_4V;
6407 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6408 int_x86_avx2_pmins_b>, VEX_4V;
6409 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6410 int_x86_avx2_pmins_d>, VEX_4V;
6411 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6412 int_x86_avx2_pminu_d>, VEX_4V;
6413 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6414 int_x86_avx2_pminu_w>, VEX_4V;
6415 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6416 int_x86_avx2_pmaxs_b>, VEX_4V;
6417 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6418 int_x86_avx2_pmaxs_d>, VEX_4V;
6419 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6420 int_x86_avx2_pmaxu_d>, VEX_4V;
6421 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6422 int_x86_avx2_pmaxu_w>, VEX_4V;
6423 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6424 int_x86_avx2_pmul_dq>, VEX_4V;
6426 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6427 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6428 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6429 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6432 let Constraints = "$src1 = $dst" in {
6433 let isCommutable = 0 in
6434 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6435 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6436 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6437 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6438 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6439 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6440 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6441 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6442 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6443 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6444 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6447 let Predicates = [HasSSE41] in {
6448 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6449 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6450 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6451 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6454 /// SS48I_binop_rm - Simple SSE41 binary operator.
6455 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6456 ValueType OpVT, bit Is2Addr = 1> {
6457 let isCommutable = 1 in
6458 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6459 (ins VR128:$src1, VR128:$src2),
6461 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6462 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6463 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6465 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6466 (ins VR128:$src1, i128mem:$src2),
6468 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6469 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6470 [(set VR128:$dst, (OpNode VR128:$src1,
6471 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6475 /// SS48I_binop_rm - Simple SSE41 binary operator.
6476 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6478 let isCommutable = 1 in
6479 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6480 (ins VR256:$src1, VR256:$src2),
6481 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6482 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6484 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6485 (ins VR256:$src1, i256mem:$src2),
6486 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6487 [(set VR256:$dst, (OpNode VR256:$src1,
6488 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6492 let Predicates = [HasAVX] in
6493 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6494 let Predicates = [HasAVX2] in
6495 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6496 let Constraints = "$src1 = $dst" in
6497 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6499 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6500 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6501 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6502 X86MemOperand x86memop, bit Is2Addr = 1> {
6503 let isCommutable = 1 in
6504 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6505 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6507 !strconcat(OpcodeStr,
6508 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6509 !strconcat(OpcodeStr,
6510 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6511 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6513 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6514 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6516 !strconcat(OpcodeStr,
6517 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6518 !strconcat(OpcodeStr,
6519 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6522 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6526 let Predicates = [HasAVX] in {
6527 let isCommutable = 0 in {
6528 let ExeDomain = SSEPackedSingle in {
6529 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6530 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6531 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6532 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6534 let ExeDomain = SSEPackedDouble in {
6535 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6536 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6537 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6538 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6540 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6541 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6542 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6543 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6545 let ExeDomain = SSEPackedSingle in
6546 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6547 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6548 let ExeDomain = SSEPackedDouble in
6549 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6550 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6551 let ExeDomain = SSEPackedSingle in
6552 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6553 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6556 let Predicates = [HasAVX2] in {
6557 let isCommutable = 0 in {
6558 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6559 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6560 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6561 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6565 let Constraints = "$src1 = $dst" in {
6566 let isCommutable = 0 in {
6567 let ExeDomain = SSEPackedSingle in
6568 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6569 VR128, memopv4f32, i128mem>;
6570 let ExeDomain = SSEPackedDouble in
6571 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6572 VR128, memopv2f64, i128mem>;
6573 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6574 VR128, memopv2i64, i128mem>;
6575 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6576 VR128, memopv2i64, i128mem>;
6578 let ExeDomain = SSEPackedSingle in
6579 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6580 VR128, memopv4f32, i128mem>;
6581 let ExeDomain = SSEPackedDouble in
6582 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6583 VR128, memopv2f64, i128mem>;
6586 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6587 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6588 RegisterClass RC, X86MemOperand x86memop,
6589 PatFrag mem_frag, Intrinsic IntId> {
6590 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6591 (ins RC:$src1, RC:$src2, RC:$src3),
6592 !strconcat(OpcodeStr,
6593 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6594 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6595 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6597 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6598 (ins RC:$src1, x86memop:$src2, RC:$src3),
6599 !strconcat(OpcodeStr,
6600 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6602 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6604 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6607 let Predicates = [HasAVX] in {
6608 let ExeDomain = SSEPackedDouble in {
6609 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6610 memopv2f64, int_x86_sse41_blendvpd>;
6611 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6612 memopv4f64, int_x86_avx_blendv_pd_256>;
6613 } // ExeDomain = SSEPackedDouble
6614 let ExeDomain = SSEPackedSingle in {
6615 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6616 memopv4f32, int_x86_sse41_blendvps>;
6617 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6618 memopv8f32, int_x86_avx_blendv_ps_256>;
6619 } // ExeDomain = SSEPackedSingle
6620 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6621 memopv2i64, int_x86_sse41_pblendvb>;
6624 let Predicates = [HasAVX2] in {
6625 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6626 memopv4i64, int_x86_avx2_pblendvb>;
6629 let Predicates = [HasAVX] in {
6630 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6631 (v16i8 VR128:$src2))),
6632 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6633 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6634 (v4i32 VR128:$src2))),
6635 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6636 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6637 (v4f32 VR128:$src2))),
6638 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6639 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6640 (v2i64 VR128:$src2))),
6641 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6642 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6643 (v2f64 VR128:$src2))),
6644 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6645 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6646 (v8i32 VR256:$src2))),
6647 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6648 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6649 (v8f32 VR256:$src2))),
6650 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6651 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6652 (v4i64 VR256:$src2))),
6653 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6654 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6655 (v4f64 VR256:$src2))),
6656 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6659 let Predicates = [HasAVX2] in {
6660 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6661 (v32i8 VR256:$src2))),
6662 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6665 /// SS41I_ternary_int - SSE 4.1 ternary operator
6666 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6667 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6669 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6670 (ins VR128:$src1, VR128:$src2),
6671 !strconcat(OpcodeStr,
6672 "\t{$src2, $dst|$dst, $src2}"),
6673 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6676 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6677 (ins VR128:$src1, i128mem:$src2),
6678 !strconcat(OpcodeStr,
6679 "\t{$src2, $dst|$dst, $src2}"),
6682 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6686 let ExeDomain = SSEPackedDouble in
6687 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6688 int_x86_sse41_blendvpd>;
6689 let ExeDomain = SSEPackedSingle in
6690 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6691 int_x86_sse41_blendvps>;
6692 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6693 int_x86_sse41_pblendvb>;
6695 let Predicates = [HasSSE41] in {
6696 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6697 (v16i8 VR128:$src2))),
6698 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6699 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6700 (v4i32 VR128:$src2))),
6701 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6702 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6703 (v4f32 VR128:$src2))),
6704 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6705 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6706 (v2i64 VR128:$src2))),
6707 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6708 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6709 (v2f64 VR128:$src2))),
6710 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6713 let Predicates = [HasAVX] in
6714 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6715 "vmovntdqa\t{$src, $dst|$dst, $src}",
6716 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6718 let Predicates = [HasAVX2] in
6719 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6720 "vmovntdqa\t{$src, $dst|$dst, $src}",
6721 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6723 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6724 "movntdqa\t{$src, $dst|$dst, $src}",
6725 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6728 //===----------------------------------------------------------------------===//
6729 // SSE4.2 - Compare Instructions
6730 //===----------------------------------------------------------------------===//
6732 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6733 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6734 Intrinsic IntId128, bit Is2Addr = 1> {
6735 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6736 (ins VR128:$src1, VR128:$src2),
6738 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6739 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6740 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6742 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6743 (ins VR128:$src1, i128mem:$src2),
6745 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6746 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6748 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6751 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6752 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6753 Intrinsic IntId256> {
6754 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6755 (ins VR256:$src1, VR256:$src2),
6756 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6757 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6759 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6760 (ins VR256:$src1, i256mem:$src2),
6761 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6763 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6766 let Predicates = [HasAVX] in {
6767 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6770 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6771 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6772 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6773 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6776 let Predicates = [HasAVX2] in {
6777 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6780 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6781 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6782 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6783 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6786 let Constraints = "$src1 = $dst" in
6787 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6789 let Predicates = [HasSSE42] in {
6790 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6791 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6792 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6793 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6796 //===----------------------------------------------------------------------===//
6797 // SSE4.2 - String/text Processing Instructions
6798 //===----------------------------------------------------------------------===//
6800 // Packed Compare Implicit Length Strings, Return Mask
6801 multiclass pseudo_pcmpistrm<string asm> {
6802 def REG : PseudoI<(outs VR128:$dst),
6803 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6804 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6806 def MEM : PseudoI<(outs VR128:$dst),
6807 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6808 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6809 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6812 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6813 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6814 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6817 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6818 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6819 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6820 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6822 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6823 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6824 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6827 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6828 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6829 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6830 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6832 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6833 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6834 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6837 // Packed Compare Explicit Length Strings, Return Mask
6838 multiclass pseudo_pcmpestrm<string asm> {
6839 def REG : PseudoI<(outs VR128:$dst),
6840 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6841 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6842 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6843 def MEM : PseudoI<(outs VR128:$dst),
6844 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6845 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6846 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6849 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6850 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6851 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6854 let Predicates = [HasAVX],
6855 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6856 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6857 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6858 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6860 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6861 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6862 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6865 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6866 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6867 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6868 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6870 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6871 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6872 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6875 // Packed Compare Implicit Length Strings, Return Index
6876 let Defs = [ECX, EFLAGS] in {
6877 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6878 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6879 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6880 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6881 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6882 (implicit EFLAGS)]>, OpSize;
6883 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6884 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6885 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6886 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6887 (implicit EFLAGS)]>, OpSize;
6891 let Predicates = [HasAVX] in {
6892 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6894 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6896 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6898 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6900 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6902 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6906 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6907 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6908 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6909 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6910 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6911 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6913 // Packed Compare Explicit Length Strings, Return Index
6914 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6915 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6916 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6917 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6918 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6919 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6920 (implicit EFLAGS)]>, OpSize;
6921 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6922 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6923 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6925 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6926 (implicit EFLAGS)]>, OpSize;
6930 let Predicates = [HasAVX] in {
6931 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6933 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6935 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6937 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6939 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6941 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6945 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6946 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6947 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6948 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6949 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6950 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6952 //===----------------------------------------------------------------------===//
6953 // SSE4.2 - CRC Instructions
6954 //===----------------------------------------------------------------------===//
6956 // No CRC instructions have AVX equivalents
6958 // crc intrinsic instruction
6959 // This set of instructions are only rm, the only difference is the size
6961 let Constraints = "$src1 = $dst" in {
6962 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6963 (ins GR32:$src1, i8mem:$src2),
6964 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6966 (int_x86_sse42_crc32_32_8 GR32:$src1,
6967 (load addr:$src2)))]>;
6968 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6969 (ins GR32:$src1, GR8:$src2),
6970 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6972 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6973 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6974 (ins GR32:$src1, i16mem:$src2),
6975 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6977 (int_x86_sse42_crc32_32_16 GR32:$src1,
6978 (load addr:$src2)))]>,
6980 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6981 (ins GR32:$src1, GR16:$src2),
6982 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6986 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6987 (ins GR32:$src1, i32mem:$src2),
6988 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_32_32 GR32:$src1,
6991 (load addr:$src2)))]>;
6992 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6993 (ins GR32:$src1, GR32:$src2),
6994 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6996 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6997 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6998 (ins GR64:$src1, i8mem:$src2),
6999 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7001 (int_x86_sse42_crc32_64_8 GR64:$src1,
7002 (load addr:$src2)))]>,
7004 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7005 (ins GR64:$src1, GR8:$src2),
7006 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7008 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7010 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7011 (ins GR64:$src1, i64mem:$src2),
7012 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7014 (int_x86_sse42_crc32_64_64 GR64:$src1,
7015 (load addr:$src2)))]>,
7017 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7018 (ins GR64:$src1, GR64:$src2),
7019 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7021 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7025 //===----------------------------------------------------------------------===//
7026 // AES-NI Instructions
7027 //===----------------------------------------------------------------------===//
7029 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7030 Intrinsic IntId128, bit Is2Addr = 1> {
7031 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7032 (ins VR128:$src1, VR128:$src2),
7034 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7035 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7036 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7038 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7039 (ins VR128:$src1, i128mem:$src2),
7041 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7042 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7044 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7047 // Perform One Round of an AES Encryption/Decryption Flow
7048 let Predicates = [HasAVX, HasAES] in {
7049 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7050 int_x86_aesni_aesenc, 0>, VEX_4V;
7051 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7052 int_x86_aesni_aesenclast, 0>, VEX_4V;
7053 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7054 int_x86_aesni_aesdec, 0>, VEX_4V;
7055 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7056 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7059 let Constraints = "$src1 = $dst" in {
7060 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7061 int_x86_aesni_aesenc>;
7062 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7063 int_x86_aesni_aesenclast>;
7064 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7065 int_x86_aesni_aesdec>;
7066 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7067 int_x86_aesni_aesdeclast>;
7070 // Perform the AES InvMixColumn Transformation
7071 let Predicates = [HasAVX, HasAES] in {
7072 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7074 "vaesimc\t{$src1, $dst|$dst, $src1}",
7076 (int_x86_aesni_aesimc VR128:$src1))]>,
7078 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7079 (ins i128mem:$src1),
7080 "vaesimc\t{$src1, $dst|$dst, $src1}",
7081 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7084 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7086 "aesimc\t{$src1, $dst|$dst, $src1}",
7088 (int_x86_aesni_aesimc VR128:$src1))]>,
7090 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7091 (ins i128mem:$src1),
7092 "aesimc\t{$src1, $dst|$dst, $src1}",
7093 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7096 // AES Round Key Generation Assist
7097 let Predicates = [HasAVX, HasAES] in {
7098 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7099 (ins VR128:$src1, i8imm:$src2),
7100 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7102 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7104 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7105 (ins i128mem:$src1, i8imm:$src2),
7106 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7108 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7111 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7112 (ins VR128:$src1, i8imm:$src2),
7113 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7115 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7117 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7118 (ins i128mem:$src1, i8imm:$src2),
7119 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7121 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7124 //===----------------------------------------------------------------------===//
7125 // CLMUL Instructions
7126 //===----------------------------------------------------------------------===//
7128 // Carry-less Multiplication instructions
7129 let neverHasSideEffects = 1 in {
7130 // AVX carry-less Multiplication instructions
7131 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7132 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7133 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7137 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7138 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7139 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7142 let Constraints = "$src1 = $dst" in {
7143 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7144 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7145 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7149 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7150 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7151 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7153 } // Constraints = "$src1 = $dst"
7154 } // neverHasSideEffects = 1
7157 multiclass pclmul_alias<string asm, int immop> {
7158 def : InstAlias<!strconcat("pclmul", asm,
7159 "dq {$src, $dst|$dst, $src}"),
7160 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7162 def : InstAlias<!strconcat("pclmul", asm,
7163 "dq {$src, $dst|$dst, $src}"),
7164 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7166 def : InstAlias<!strconcat("vpclmul", asm,
7167 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7168 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7170 def : InstAlias<!strconcat("vpclmul", asm,
7171 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7172 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7174 defm : pclmul_alias<"hqhq", 0x11>;
7175 defm : pclmul_alias<"hqlq", 0x01>;
7176 defm : pclmul_alias<"lqhq", 0x10>;
7177 defm : pclmul_alias<"lqlq", 0x00>;
7179 //===----------------------------------------------------------------------===//
7181 //===----------------------------------------------------------------------===//
7183 //===----------------------------------------------------------------------===//
7184 // VBROADCAST - Load from memory and broadcast to all elements of the
7185 // destination operand
7187 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7188 X86MemOperand x86memop, Intrinsic Int> :
7189 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7190 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7191 [(set RC:$dst, (Int addr:$src))]>, VEX;
7193 // AVX2 adds register forms
7194 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7196 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7197 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7198 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7200 let ExeDomain = SSEPackedSingle in {
7201 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7202 int_x86_avx_vbroadcast_ss>;
7203 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7204 int_x86_avx_vbroadcast_ss_256>;
7206 let ExeDomain = SSEPackedDouble in
7207 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7208 int_x86_avx_vbroadcast_sd_256>;
7209 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7210 int_x86_avx_vbroadcastf128_pd_256>;
7212 let ExeDomain = SSEPackedSingle in {
7213 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7214 int_x86_avx2_vbroadcast_ss_ps>;
7215 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7216 int_x86_avx2_vbroadcast_ss_ps_256>;
7218 let ExeDomain = SSEPackedDouble in
7219 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7220 int_x86_avx2_vbroadcast_sd_pd_256>;
7222 let Predicates = [HasAVX2] in
7223 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7224 int_x86_avx2_vbroadcasti128>;
7226 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7227 (VBROADCASTF128 addr:$src)>;
7230 //===----------------------------------------------------------------------===//
7231 // VINSERTF128 - Insert packed floating-point values
7233 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7234 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7235 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7236 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7239 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7240 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7241 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7245 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7246 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7247 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7248 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7249 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7250 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7252 //===----------------------------------------------------------------------===//
7253 // VEXTRACTF128 - Extract packed floating-point values
7255 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7256 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7257 (ins VR256:$src1, i8imm:$src2),
7258 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7261 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7262 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7263 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7267 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7268 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7269 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7270 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7271 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7272 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7274 //===----------------------------------------------------------------------===//
7275 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7277 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7278 Intrinsic IntLd, Intrinsic IntLd256,
7279 Intrinsic IntSt, Intrinsic IntSt256> {
7280 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7281 (ins VR128:$src1, f128mem:$src2),
7282 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7283 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7285 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7286 (ins VR256:$src1, f256mem:$src2),
7287 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7288 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7290 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7291 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7292 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7293 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7294 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7295 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7296 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7297 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7300 let ExeDomain = SSEPackedSingle in
7301 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7302 int_x86_avx_maskload_ps,
7303 int_x86_avx_maskload_ps_256,
7304 int_x86_avx_maskstore_ps,
7305 int_x86_avx_maskstore_ps_256>;
7306 let ExeDomain = SSEPackedDouble in
7307 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7308 int_x86_avx_maskload_pd,
7309 int_x86_avx_maskload_pd_256,
7310 int_x86_avx_maskstore_pd,
7311 int_x86_avx_maskstore_pd_256>;
7313 //===----------------------------------------------------------------------===//
7314 // VPERMIL - Permute Single and Double Floating-Point Values
7316 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7317 RegisterClass RC, X86MemOperand x86memop_f,
7318 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7319 Intrinsic IntVar, Intrinsic IntImm> {
7320 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7321 (ins RC:$src1, RC:$src2),
7322 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7323 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7324 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7325 (ins RC:$src1, x86memop_i:$src2),
7326 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7327 [(set RC:$dst, (IntVar RC:$src1,
7328 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7330 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7331 (ins RC:$src1, i8imm:$src2),
7332 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7333 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7334 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7335 (ins x86memop_f:$src1, i8imm:$src2),
7336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7337 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7340 let ExeDomain = SSEPackedSingle in {
7341 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7342 memopv4f32, memopv2i64,
7343 int_x86_avx_vpermilvar_ps,
7344 int_x86_avx_vpermil_ps>;
7345 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7346 memopv8f32, memopv4i64,
7347 int_x86_avx_vpermilvar_ps_256,
7348 int_x86_avx_vpermil_ps_256>;
7350 let ExeDomain = SSEPackedDouble in {
7351 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7352 memopv2f64, memopv2i64,
7353 int_x86_avx_vpermilvar_pd,
7354 int_x86_avx_vpermil_pd>;
7355 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7356 memopv4f64, memopv4i64,
7357 int_x86_avx_vpermilvar_pd_256,
7358 int_x86_avx_vpermil_pd_256>;
7361 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7362 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7363 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7364 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7365 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7366 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7367 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7368 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7369 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7370 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7371 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7372 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7373 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7375 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7376 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7377 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7379 //===----------------------------------------------------------------------===//
7380 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7382 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7383 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7384 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7385 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7388 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7389 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7390 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7394 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7395 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7396 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7397 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7398 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7399 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7401 def : Pat<(int_x86_avx_vperm2f128_ps_256
7402 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7403 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7404 def : Pat<(int_x86_avx_vperm2f128_pd_256
7405 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7406 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7407 def : Pat<(int_x86_avx_vperm2f128_si_256
7408 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7409 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7411 //===----------------------------------------------------------------------===//
7412 // VZERO - Zero YMM registers
7414 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7415 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7416 // Zero All YMM registers
7417 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7418 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7420 // Zero Upper bits of YMM registers
7421 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7422 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7425 //===----------------------------------------------------------------------===//
7426 // Half precision conversion instructions
7427 //===----------------------------------------------------------------------===//
7428 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7429 let Predicates = [HasAVX, HasF16C] in {
7430 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7431 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7432 [(set RC:$dst, (Int VR128:$src))]>,
7434 let neverHasSideEffects = 1, mayLoad = 1 in
7435 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7436 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7440 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7441 let Predicates = [HasAVX, HasF16C] in {
7442 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7443 (ins RC:$src1, i32i8imm:$src2),
7444 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7445 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7447 let neverHasSideEffects = 1, mayLoad = 1 in
7448 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7449 (ins RC:$src1, i32i8imm:$src2),
7450 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7455 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7456 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7457 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7458 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7460 //===----------------------------------------------------------------------===//
7461 // AVX2 Instructions
7462 //===----------------------------------------------------------------------===//
7464 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7465 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7466 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7467 X86MemOperand x86memop> {
7468 let isCommutable = 1 in
7469 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7470 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7471 !strconcat(OpcodeStr,
7472 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7473 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7475 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7476 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7477 !strconcat(OpcodeStr,
7478 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7481 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7485 let isCommutable = 0 in {
7486 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7487 VR128, memopv2i64, i128mem>;
7488 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7489 VR256, memopv4i64, i256mem>;
7492 //===----------------------------------------------------------------------===//
7493 // VPBROADCAST - Load from memory and broadcast to all elements of the
7494 // destination operand
7496 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7497 X86MemOperand x86memop, PatFrag ld_frag,
7498 Intrinsic Int128, Intrinsic Int256> {
7499 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7500 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7501 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7502 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7505 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7506 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7507 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7508 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7509 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7510 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7512 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7515 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7516 int_x86_avx2_pbroadcastb_128,
7517 int_x86_avx2_pbroadcastb_256>;
7518 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7519 int_x86_avx2_pbroadcastw_128,
7520 int_x86_avx2_pbroadcastw_256>;
7521 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7522 int_x86_avx2_pbroadcastd_128,
7523 int_x86_avx2_pbroadcastd_256>;
7524 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7525 int_x86_avx2_pbroadcastq_128,
7526 int_x86_avx2_pbroadcastq_256>;
7528 let Predicates = [HasAVX2] in {
7529 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7530 (VPBROADCASTBrm addr:$src)>;
7531 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7532 (VPBROADCASTBYrm addr:$src)>;
7533 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7534 (VPBROADCASTWrm addr:$src)>;
7535 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7536 (VPBROADCASTWYrm addr:$src)>;
7537 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7538 (VPBROADCASTDrm addr:$src)>;
7539 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7540 (VPBROADCASTDYrm addr:$src)>;
7541 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7542 (VPBROADCASTQrm addr:$src)>;
7543 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7544 (VPBROADCASTQYrm addr:$src)>;
7547 // AVX1 broadcast patterns
7548 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7549 (VBROADCASTSSYrm addr:$src)>;
7550 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7551 (VBROADCASTSDrm addr:$src)>;
7552 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7553 (VBROADCASTSSYrm addr:$src)>;
7554 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7555 (VBROADCASTSDrm addr:$src)>;
7557 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7558 (VBROADCASTSSrm addr:$src)>;
7559 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7560 (VBROADCASTSSrm addr:$src)>;
7562 //===----------------------------------------------------------------------===//
7563 // VPERM - Permute instructions
7566 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7568 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7569 (ins VR256:$src1, VR256:$src2),
7570 !strconcat(OpcodeStr,
7571 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7572 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7573 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7574 (ins VR256:$src1, i256mem:$src2),
7575 !strconcat(OpcodeStr,
7576 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7577 [(set VR256:$dst, (Int VR256:$src1,
7578 (bitconvert (mem_frag addr:$src2))))]>,
7582 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7583 let ExeDomain = SSEPackedSingle in
7584 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7586 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7588 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7589 (ins VR256:$src1, i8imm:$src2),
7590 !strconcat(OpcodeStr,
7591 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7592 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7593 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7594 (ins i256mem:$src1, i8imm:$src2),
7595 !strconcat(OpcodeStr,
7596 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7597 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7601 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7603 let ExeDomain = SSEPackedDouble in
7604 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7607 //===----------------------------------------------------------------------===//
7608 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7610 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7611 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7612 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7614 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7616 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7617 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7618 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7620 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7624 let Predicates = [HasAVX2] in {
7625 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7626 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7627 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7628 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7629 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7630 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7631 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7632 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7634 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7636 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7637 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7638 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7639 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7640 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7642 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7643 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7645 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7649 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7650 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7651 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7652 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7653 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7654 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7655 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7656 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7657 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7658 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7659 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7660 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7662 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7663 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7664 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7665 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7666 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7667 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7668 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7669 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7670 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7671 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7672 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7673 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7674 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7675 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7676 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7677 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7678 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7679 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7682 //===----------------------------------------------------------------------===//
7683 // VINSERTI128 - Insert packed integer values
7685 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7686 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7687 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7689 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7691 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7692 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7693 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7695 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7696 imm:$src3))]>, VEX_4V;
7698 let Predicates = [HasAVX2] in {
7699 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7701 (VINSERTI128rr VR256:$src1, VR128:$src2,
7702 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7703 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7705 (VINSERTI128rr VR256:$src1, VR128:$src2,
7706 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7707 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7709 (VINSERTI128rr VR256:$src1, VR128:$src2,
7710 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7711 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7713 (VINSERTI128rr VR256:$src1, VR128:$src2,
7714 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7718 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7720 (VINSERTF128rr VR256:$src1, VR128:$src2,
7721 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7722 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7724 (VINSERTF128rr VR256:$src1, VR128:$src2,
7725 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7726 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7728 (VINSERTF128rr VR256:$src1, VR128:$src2,
7729 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7730 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7732 (VINSERTF128rr VR256:$src1, VR128:$src2,
7733 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7734 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7736 (VINSERTF128rr VR256:$src1, VR128:$src2,
7737 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7738 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7740 (VINSERTF128rr VR256:$src1, VR128:$src2,
7741 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7743 //===----------------------------------------------------------------------===//
7744 // VEXTRACTI128 - Extract packed integer values
7746 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7747 (ins VR256:$src1, i8imm:$src2),
7748 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7750 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7752 let neverHasSideEffects = 1, mayStore = 1 in
7753 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7754 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7755 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7757 let Predicates = [HasAVX2] in {
7758 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7759 (v2i64 (VEXTRACTI128rr
7760 (v4i64 VR256:$src1),
7761 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7762 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7763 (v4i32 (VEXTRACTI128rr
7764 (v8i32 VR256:$src1),
7765 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7766 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7767 (v8i16 (VEXTRACTI128rr
7768 (v16i16 VR256:$src1),
7769 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7770 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7771 (v16i8 (VEXTRACTI128rr
7772 (v32i8 VR256:$src1),
7773 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7777 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7778 (v4f32 (VEXTRACTF128rr
7779 (v8f32 VR256:$src1),
7780 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7781 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7782 (v2f64 (VEXTRACTF128rr
7783 (v4f64 VR256:$src1),
7784 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7785 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7786 (v2i64 (VEXTRACTF128rr
7787 (v4i64 VR256:$src1),
7788 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7789 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7790 (v4i32 (VEXTRACTF128rr
7791 (v8i32 VR256:$src1),
7792 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7793 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7794 (v8i16 (VEXTRACTF128rr
7795 (v16i16 VR256:$src1),
7796 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7797 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7798 (v16i8 (VEXTRACTF128rr
7799 (v32i8 VR256:$src1),
7800 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7802 //===----------------------------------------------------------------------===//
7803 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7805 multiclass avx2_pmovmask<string OpcodeStr,
7806 Intrinsic IntLd128, Intrinsic IntLd256,
7807 Intrinsic IntSt128, Intrinsic IntSt256> {
7808 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7809 (ins VR128:$src1, i128mem:$src2),
7810 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7811 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7812 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7813 (ins VR256:$src1, i256mem:$src2),
7814 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7815 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7816 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7817 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7818 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7819 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7820 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7821 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7822 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7823 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7826 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7827 int_x86_avx2_maskload_d,
7828 int_x86_avx2_maskload_d_256,
7829 int_x86_avx2_maskstore_d,
7830 int_x86_avx2_maskstore_d_256>;
7831 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7832 int_x86_avx2_maskload_q,
7833 int_x86_avx2_maskload_q_256,
7834 int_x86_avx2_maskstore_q,
7835 int_x86_avx2_maskstore_q_256>, VEX_W;
7838 //===----------------------------------------------------------------------===//
7839 // Variable Bit Shifts
7841 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7842 ValueType vt128, ValueType vt256> {
7843 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7844 (ins VR128:$src1, VR128:$src2),
7845 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7847 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7849 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7850 (ins VR128:$src1, i128mem:$src2),
7851 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7853 (vt128 (OpNode VR128:$src1,
7854 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7856 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7857 (ins VR256:$src1, VR256:$src2),
7858 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7860 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7862 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7863 (ins VR256:$src1, i256mem:$src2),
7864 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7866 (vt256 (OpNode VR256:$src1,
7867 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7871 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7872 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7873 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7874 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7875 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;