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:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 // AVX has no support for 256-bit integer instructions, but since the 128-bit
291 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
292 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
293 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
294 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
296 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
297 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
298 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
300 // We set canFoldAsLoad because this can be converted to a constant-pool
301 // load of an all-ones value if folding it would be beneficial.
302 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
303 // JIT implementation, it does not expand the instructions below like
304 // X86MCInstLower does.
305 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
306 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
307 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
308 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
309 let Predicates = [HasAVX] in
310 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
311 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
312 let Predicates = [HasAVX2] in
313 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$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:$dst)]>;
407 let Predicates = [HasSSE1] in {
408 let AddedComplexity = 15 in {
409 // Extract the low 32-bit value from one vector and insert it into another.
410 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
411 (MOVSSrr (v4f32 VR128:$src1),
412 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
413 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
414 (MOVSSrr (v4i32 VR128:$src1),
415 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
417 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
418 // MOVSS to the lower bits.
419 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
420 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
421 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
422 (MOVSSrr (v4f32 (V_SET0)),
423 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
424 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
425 (MOVSSrr (v4i32 (V_SET0)),
426 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
429 let AddedComplexity = 20 in {
430 // MOVSSrm zeros the high parts of the register; represent this
431 // with SUBREG_TO_REG.
432 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
433 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
434 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
435 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
436 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
437 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
440 // Extract and store.
441 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
444 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
446 // Shuffle with MOVSS
447 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
448 (MOVSSrr VR128:$src1, FR32:$src2)>;
449 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
450 (MOVSSrr (v4i32 VR128:$src1),
451 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
452 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
453 (MOVSSrr (v4f32 VR128:$src1),
454 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
457 let Predicates = [HasSSE2] in {
458 let AddedComplexity = 15 in {
459 // Extract the low 64-bit value from one vector and insert it into another.
460 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
461 (MOVSDrr (v2f64 VR128:$src1),
462 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
463 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
464 (MOVSDrr (v2i64 VR128:$src1),
465 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
467 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
468 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
469 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
470 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
471 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
473 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
474 // MOVSD to the lower bits.
475 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
476 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
479 let AddedComplexity = 20 in {
480 // MOVSDrm zeros the high parts of the register; represent this
481 // with SUBREG_TO_REG.
482 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
483 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
484 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
485 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
486 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
487 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
488 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
489 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
490 def : Pat<(v2f64 (X86vzload addr:$src)),
491 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
494 // Extract and store.
495 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
498 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
500 // Shuffle with MOVSD
501 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
502 (MOVSDrr VR128:$src1, FR64:$src2)>;
503 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
504 (MOVSDrr (v2i64 VR128:$src1),
505 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
506 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
507 (MOVSDrr (v2f64 VR128:$src1),
508 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
509 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
510 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
511 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
512 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
514 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
515 // is during lowering, where it's not possible to recognize the fold cause
516 // it has two uses through a bitcast. One use disappears at isel time and the
517 // fold opportunity reappears.
518 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
519 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
520 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
521 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
522 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
523 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
524 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
525 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
528 let Predicates = [HasAVX] in {
529 let AddedComplexity = 15 in {
530 // Extract the low 32-bit value from one vector and insert it into another.
531 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
532 (VMOVSSrr (v4f32 VR128:$src1),
533 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
534 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
535 (VMOVSSrr (v4i32 VR128:$src1),
536 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
538 // Extract the low 64-bit value from one vector and insert it into another.
539 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
540 (VMOVSDrr (v2f64 VR128:$src1),
541 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
542 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
543 (VMOVSDrr (v2i64 VR128:$src1),
544 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
546 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
547 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
548 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
549 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
550 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
552 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
553 // MOVS{S,D} to the lower bits.
554 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
555 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
556 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
557 (VMOVSSrr (v4f32 (V_SET0)),
558 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
559 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
560 (VMOVSSrr (v4i32 (V_SET0)),
561 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
562 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
563 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
565 // Move low f32 and clear high bits.
566 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
567 (SUBREG_TO_REG (i32 0),
568 (VMOVSSrr (v4f32 (V_SET0)),
569 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
570 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
571 (SUBREG_TO_REG (i32 0),
572 (VMOVSSrr (v4i32 (V_SET0)),
573 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
576 let AddedComplexity = 20 in {
577 // MOVSSrm zeros the high parts of the register; represent this
578 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
579 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
581 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
582 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
583 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
584 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
586 // MOVSDrm zeros the high parts of the register; represent this
587 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
588 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
589 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
590 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
591 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
592 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
593 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
594 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
595 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
596 def : Pat<(v2f64 (X86vzload addr:$src)),
597 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
599 // Represent the same patterns above but in the form they appear for
601 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
602 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
603 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
604 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
605 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
606 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
607 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
608 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
609 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
611 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
612 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
613 (SUBREG_TO_REG (i32 0),
614 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
616 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
617 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
618 (SUBREG_TO_REG (i64 0),
619 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
621 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
622 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
623 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
625 // Move low f64 and clear high bits.
626 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
627 (SUBREG_TO_REG (i32 0),
628 (VMOVSDrr (v2f64 (V_SET0)),
629 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
631 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
632 (SUBREG_TO_REG (i32 0),
633 (VMOVSDrr (v2i64 (V_SET0)),
634 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
636 // Extract and store.
637 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
640 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
641 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
644 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
646 // Shuffle with VMOVSS
647 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
648 (VMOVSSrr VR128:$src1, FR32:$src2)>;
649 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
650 (VMOVSSrr (v4i32 VR128:$src1),
651 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
652 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
653 (VMOVSSrr (v4f32 VR128:$src1),
654 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
657 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
658 (SUBREG_TO_REG (i32 0),
659 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
660 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
661 def : Pat<(v8f32 (X86Movsd VR256:$src1, VR256:$src2)),
662 (SUBREG_TO_REG (i32 0),
663 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
664 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
666 // Shuffle with VMOVSD
667 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
668 (VMOVSDrr VR128:$src1, FR64:$src2)>;
669 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
670 (VMOVSDrr (v2i64 VR128:$src1),
671 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
672 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
673 (VMOVSDrr (v2f64 VR128:$src1),
674 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
675 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
676 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
678 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
679 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
683 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
684 (SUBREG_TO_REG (i32 0),
685 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
686 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
687 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
688 (SUBREG_TO_REG (i32 0),
689 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
690 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
693 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
694 // is during lowering, where it's not possible to recognize the fold cause
695 // it has two uses through a bitcast. One use disappears at isel time and the
696 // fold opportunity reappears.
697 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
698 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
700 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
701 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
703 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
704 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
706 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
707 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
711 //===----------------------------------------------------------------------===//
712 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
713 //===----------------------------------------------------------------------===//
715 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
716 X86MemOperand x86memop, PatFrag ld_frag,
717 string asm, Domain d,
718 bit IsReMaterializable = 1> {
719 let neverHasSideEffects = 1 in
720 def rr : PI<opc, MRMSrcReg, (outs RC:$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:$src)>;
870 // Use movaps / movups for SSE integer load / store (one byte shorter).
871 // The instructions selected below are then converted to MOVDQA/MOVDQU
872 // during the SSE domain pass.
873 let Predicates = [HasSSE1] in {
874 def : Pat<(alignedloadv4i32 addr:$src),
875 (MOVAPSrm addr:$src)>;
876 def : Pat<(loadv4i32 addr:$src),
877 (MOVUPSrm addr:$src)>;
878 def : Pat<(alignedloadv2i64 addr:$src),
879 (MOVAPSrm addr:$src)>;
880 def : Pat<(loadv2i64 addr:$src),
881 (MOVUPSrm addr:$src)>;
883 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
884 (MOVAPSmr addr:$dst, VR128:$src)>;
885 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
886 (MOVAPSmr addr:$dst, VR128:$src)>;
887 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
888 (MOVAPSmr addr:$dst, VR128:$src)>;
889 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
890 (MOVAPSmr addr:$dst, VR128:$src)>;
891 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
892 (MOVUPSmr addr:$dst, VR128:$src)>;
893 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
894 (MOVUPSmr addr:$dst, VR128:$src)>;
895 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
896 (MOVUPSmr addr:$dst, VR128:$src)>;
897 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
898 (MOVUPSmr addr:$dst, VR128:$src)>;
901 // Use vmovaps/vmovups for AVX integer load/store.
902 let Predicates = [HasAVX] in {
903 // 128-bit load/store
904 def : Pat<(alignedloadv4i32 addr:$src),
905 (VMOVAPSrm addr:$src)>;
906 def : Pat<(loadv4i32 addr:$src),
907 (VMOVUPSrm addr:$src)>;
908 def : Pat<(alignedloadv2i64 addr:$src),
909 (VMOVAPSrm addr:$src)>;
910 def : Pat<(loadv2i64 addr:$src),
911 (VMOVUPSrm addr:$src)>;
913 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
914 (VMOVAPSmr addr:$dst, VR128:$src)>;
915 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
916 (VMOVAPSmr addr:$dst, VR128:$src)>;
917 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
918 (VMOVAPSmr addr:$dst, VR128:$src)>;
919 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
920 (VMOVAPSmr addr:$dst, VR128:$src)>;
921 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
922 (VMOVUPSmr addr:$dst, VR128:$src)>;
923 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
924 (VMOVUPSmr addr:$dst, VR128:$src)>;
925 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
926 (VMOVUPSmr addr:$dst, VR128:$src)>;
927 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
928 (VMOVUPSmr addr:$dst, VR128:$src)>;
930 // 256-bit load/store
931 def : Pat<(alignedloadv4i64 addr:$src),
932 (VMOVAPSYrm addr:$src)>;
933 def : Pat<(loadv4i64 addr:$src),
934 (VMOVUPSYrm addr:$src)>;
935 def : Pat<(alignedloadv8i32 addr:$src),
936 (VMOVAPSYrm addr:$src)>;
937 def : Pat<(loadv8i32 addr:$src),
938 (VMOVUPSYrm addr:$src)>;
939 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
940 (VMOVAPSYmr addr:$dst, VR256:$src)>;
941 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
942 (VMOVAPSYmr addr:$dst, VR256:$src)>;
943 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
944 (VMOVAPSYmr addr:$dst, VR256:$src)>;
945 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
946 (VMOVAPSYmr addr:$dst, VR256:$src)>;
947 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
948 (VMOVUPSYmr addr:$dst, VR256:$src)>;
949 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
950 (VMOVUPSYmr addr:$dst, VR256:$src)>;
951 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
952 (VMOVUPSYmr addr:$dst, VR256:$src)>;
953 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
954 (VMOVUPSYmr addr:$dst, VR256:$src)>;
957 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
958 // bits are disregarded. FIXME: Set encoding to pseudo!
959 let neverHasSideEffects = 1 in {
960 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
961 "movaps\t{$src, $dst|$dst, $src}", []>;
962 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
963 "movapd\t{$src, $dst|$dst, $src}", []>;
964 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
965 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
966 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
967 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
970 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
971 // bits are disregarded. FIXME: Set encoding to pseudo!
972 let canFoldAsLoad = 1, isReMaterializable = 1 in {
973 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
974 "movaps\t{$src, $dst|$dst, $src}",
975 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
976 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
977 "movapd\t{$src, $dst|$dst, $src}",
978 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
979 let isCodeGenOnly = 1 in {
980 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
981 "movaps\t{$src, $dst|$dst, $src}",
982 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
983 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
984 "movapd\t{$src, $dst|$dst, $src}",
985 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
989 //===----------------------------------------------------------------------===//
990 // SSE 1 & 2 - Move Low packed FP Instructions
991 //===----------------------------------------------------------------------===//
993 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
994 PatFrag mov_frag, string base_opc,
996 def PSrm : PI<opc, MRMSrcMem,
997 (outs VR128:$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:$src2)>;
1238 def : Pat<(store (f64 (vector_extract
1239 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1240 (VMOVHPSmr addr:$dst, VR128:$src)>;
1241 def : Pat<(store (f64 (vector_extract
1242 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1243 (VMOVHPDmr addr:$dst, VR128:$src)>;
1246 let Predicates = [HasSSE1] in {
1248 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1249 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1250 def : Pat<(X86Movlhps VR128:$src1,
1251 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1252 (MOVHPSrm VR128:$src1, addr:$src2)>;
1253 def : Pat<(X86Movlhps VR128:$src1,
1254 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1255 (MOVHPSrm VR128:$src1, addr:$src2)>;
1256 def : Pat<(X86Movlhps VR128:$src1,
1257 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1258 (MOVHPSrm VR128:$src1, addr:$src2)>;
1261 def : Pat<(store (f64 (vector_extract
1262 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))), addr:$dst),
1263 (MOVHPSmr addr:$dst, VR128:$src)>;
1266 let Predicates = [HasSSE2] in {
1267 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1268 // is during lowering, where it's not possible to recognize the load fold
1269 // cause it has two uses through a bitcast. One use disappears at isel time
1270 // and the fold opportunity reappears.
1271 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1272 (scalar_to_vector (loadf64 addr:$src2)))),
1273 (MOVHPDrm VR128:$src1, addr:$src2)>;
1275 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1276 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1277 (scalar_to_vector (loadf64 addr:$src2)))),
1278 (MOVHPDrm VR128:$src1, addr:$src2)>;
1281 def : Pat<(store (f64 (vector_extract
1282 (v2f64 (X86Unpckh VR128:$src, VR128:$src)), (iPTR 0))),addr:$dst),
1283 (MOVHPDmr addr:$dst, VR128:$src)>;
1286 //===----------------------------------------------------------------------===//
1287 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1288 //===----------------------------------------------------------------------===//
1290 let AddedComplexity = 20 in {
1291 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1292 (ins VR128:$src1, VR128:$src2),
1293 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1295 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1297 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1298 (ins VR128:$src1, VR128:$src2),
1299 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1301 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1304 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1305 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1306 (ins VR128:$src1, VR128:$src2),
1307 "movlhps\t{$src2, $dst|$dst, $src2}",
1309 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1310 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1311 (ins VR128:$src1, VR128:$src2),
1312 "movhlps\t{$src2, $dst|$dst, $src2}",
1314 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1317 let Predicates = [HasAVX] in {
1319 let AddedComplexity = 20 in {
1320 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1321 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1322 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1323 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1325 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1326 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1327 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1329 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1330 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1331 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1332 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1333 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1334 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1337 let AddedComplexity = 20 in {
1338 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1339 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1340 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1342 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1343 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1344 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1345 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1349 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1350 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1351 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1352 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1355 let Predicates = [HasSSE1] in {
1357 let AddedComplexity = 20 in {
1358 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1359 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1360 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1361 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1363 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1364 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1365 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1367 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1368 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1369 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1370 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1371 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1372 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1375 let AddedComplexity = 20 in {
1376 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1377 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1378 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1380 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1381 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1382 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1383 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1384 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1387 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1388 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1389 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1390 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1393 //===----------------------------------------------------------------------===//
1394 // SSE 1 & 2 - Conversion Instructions
1395 //===----------------------------------------------------------------------===//
1397 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1398 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1400 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1401 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1402 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1403 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1406 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1407 X86MemOperand x86memop, string asm> {
1408 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1410 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1413 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1414 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1415 string asm, Domain d> {
1416 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1417 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1418 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1419 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1422 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1423 X86MemOperand x86memop, string asm> {
1424 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1425 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1427 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1428 (ins DstRC:$src1, x86memop:$src),
1429 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1432 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1433 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1435 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1436 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1438 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1439 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1441 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1442 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1443 VEX, VEX_W, VEX_LIG;
1445 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1446 // register, but the same isn't true when only using memory operands,
1447 // provide other assembly "l" and "q" forms to address this explicitly
1448 // where appropriate to do so.
1449 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1451 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1452 VEX_4V, VEX_W, VEX_LIG;
1453 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1455 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1457 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1458 VEX_4V, VEX_W, VEX_LIG;
1460 let Predicates = [HasAVX] in {
1461 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1462 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1463 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1464 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1466 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1467 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1468 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1470 def : Pat<(f32 (sint_to_fp GR32:$src)),
1471 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1472 def : Pat<(f32 (sint_to_fp GR64:$src)),
1473 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1474 def : Pat<(f64 (sint_to_fp GR32:$src)),
1475 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1476 def : Pat<(f64 (sint_to_fp GR64:$src)),
1477 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1480 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1481 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1482 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1483 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1484 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1485 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1486 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1487 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1488 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1489 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1490 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1491 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1492 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1493 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1494 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1495 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1497 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1498 // and/or XMM operand(s).
1500 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1501 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1503 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1504 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1505 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1506 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1507 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1508 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1511 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1512 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1513 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1514 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1516 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1517 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1518 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1519 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1520 (ins DstRC:$src1, x86memop:$src2),
1522 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1523 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1524 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1527 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1528 f128mem, load, "cvtsd2si">, XD, VEX;
1529 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1530 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1533 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1534 // Get rid of this hack or rename the intrinsics, there are several
1535 // intructions that only match with the intrinsic form, why create duplicates
1536 // to let them be recognized by the assembler?
1537 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1538 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1539 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1540 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1543 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1544 f128mem, load, "cvtsd2si{l}">, XD;
1545 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1546 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1549 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1550 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1551 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1552 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1554 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1555 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1556 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1557 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1560 let Constraints = "$src1 = $dst" in {
1561 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1562 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1564 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1565 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1566 "cvtsi2ss{q}">, XS, REX_W;
1567 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1568 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1570 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1571 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1572 "cvtsi2sd">, XD, REX_W;
1577 // Aliases for intrinsics
1578 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1579 f32mem, load, "cvttss2si">, XS, VEX;
1580 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1581 int_x86_sse_cvttss2si64, f32mem, load,
1582 "cvttss2si">, XS, VEX, VEX_W;
1583 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1584 f128mem, load, "cvttsd2si">, XD, VEX;
1585 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1586 int_x86_sse2_cvttsd2si64, f128mem, load,
1587 "cvttsd2si">, XD, VEX, VEX_W;
1588 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1589 f32mem, load, "cvttss2si">, XS;
1590 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1591 int_x86_sse_cvttss2si64, f32mem, load,
1592 "cvttss2si{q}">, XS, REX_W;
1593 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1594 f128mem, load, "cvttsd2si">, XD;
1595 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1596 int_x86_sse2_cvttsd2si64, f128mem, load,
1597 "cvttsd2si{q}">, XD, REX_W;
1599 let Pattern = []<dag> in {
1600 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1601 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1603 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1604 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1606 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1607 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1608 SSEPackedSingle>, TB, VEX;
1609 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1610 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1611 SSEPackedSingle>, TB, VEX;
1614 let Pattern = []<dag> in {
1615 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1616 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1617 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1618 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1619 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1620 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1621 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1624 let Predicates = [HasSSE1] in {
1625 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1626 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1627 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1628 (CVTSS2SIrm addr:$src)>;
1629 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1630 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1631 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1632 (CVTSS2SI64rm addr:$src)>;
1635 let Predicates = [HasAVX] in {
1636 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1637 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1638 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1639 (VCVTSS2SIrm addr:$src)>;
1640 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1641 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1642 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1643 (VCVTSS2SI64rm addr:$src)>;
1648 // Convert scalar double to scalar single
1649 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1650 (ins FR64:$src1, FR64:$src2),
1651 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1654 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1655 (ins FR64:$src1, f64mem:$src2),
1656 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1657 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1659 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1662 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1663 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1664 [(set FR32:$dst, (fround FR64:$src))]>;
1665 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1666 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1667 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1668 Requires<[HasSSE2, OptForSize]>;
1670 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1671 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1673 let Constraints = "$src1 = $dst" in
1674 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1675 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1677 // Convert scalar single to scalar double
1678 // SSE2 instructions with XS prefix
1679 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1680 (ins FR32:$src1, FR32:$src2),
1681 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1682 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1684 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1685 (ins FR32:$src1, f32mem:$src2),
1686 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1687 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1689 let Predicates = [HasAVX] in {
1690 def : Pat<(f64 (fextend FR32:$src)),
1691 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1692 def : Pat<(fextend (loadf32 addr:$src)),
1693 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1694 def : Pat<(extloadf32 addr:$src),
1695 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1698 def : Pat<(extloadf32 addr:$src),
1699 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1700 Requires<[HasAVX, OptForSpeed]>;
1702 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1703 "cvtss2sd\t{$src, $dst|$dst, $src}",
1704 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1705 Requires<[HasSSE2]>;
1706 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1707 "cvtss2sd\t{$src, $dst|$dst, $src}",
1708 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1709 Requires<[HasSSE2, OptForSize]>;
1711 // extload f32 -> f64. This matches load+fextend because we have a hack in
1712 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1714 // Since these loads aren't folded into the fextend, we have to match it
1716 def : Pat<(fextend (loadf32 addr:$src)),
1717 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1718 def : Pat<(extloadf32 addr:$src),
1719 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1721 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1722 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1725 VR128:$src2))]>, XS, VEX_4V,
1727 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1728 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1729 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1730 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1731 (load addr:$src2)))]>, XS, VEX_4V,
1733 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1734 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1735 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1738 VR128:$src2))]>, XS,
1739 Requires<[HasSSE2]>;
1740 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1741 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1742 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1743 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1744 (load addr:$src2)))]>, XS,
1745 Requires<[HasSSE2]>;
1748 // Convert doubleword to packed single/double fp
1749 // SSE2 instructions without OpSize prefix
1750 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1751 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1752 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1753 TB, VEX, Requires<[HasAVX]>;
1754 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1755 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1756 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1757 (bitconvert (memopv2i64 addr:$src))))]>,
1758 TB, VEX, Requires<[HasAVX]>;
1759 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1760 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1762 TB, Requires<[HasSSE2]>;
1763 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1764 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1766 (bitconvert (memopv2i64 addr:$src))))]>,
1767 TB, Requires<[HasSSE2]>;
1769 // FIXME: why the non-intrinsic version is described as SSE3?
1770 // SSE2 instructions with XS prefix
1771 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1772 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1773 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1774 XS, VEX, Requires<[HasAVX]>;
1775 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1776 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1777 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1778 (bitconvert (memopv2i64 addr:$src))))]>,
1779 XS, VEX, Requires<[HasAVX]>;
1780 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1781 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1782 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1783 XS, Requires<[HasSSE2]>;
1784 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1785 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1786 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1787 (bitconvert (memopv2i64 addr:$src))))]>,
1788 XS, Requires<[HasSSE2]>;
1791 // Convert packed single/double fp to doubleword
1792 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1794 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1796 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1798 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1800 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1802 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1803 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1805 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1806 "cvtps2dq\t{$src, $dst|$dst, $src}",
1807 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1809 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1811 "cvtps2dq\t{$src, $dst|$dst, $src}",
1812 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1813 (memop addr:$src)))]>, VEX;
1814 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1815 "cvtps2dq\t{$src, $dst|$dst, $src}",
1816 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1817 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1818 "cvtps2dq\t{$src, $dst|$dst, $src}",
1819 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1820 (memop addr:$src)))]>;
1822 // SSE2 packed instructions with XD prefix
1823 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1825 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1826 XD, VEX, Requires<[HasAVX]>;
1827 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1828 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1829 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1830 (memop addr:$src)))]>,
1831 XD, VEX, Requires<[HasAVX]>;
1832 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1833 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1834 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1835 XD, Requires<[HasSSE2]>;
1836 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1837 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1838 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1839 (memop addr:$src)))]>,
1840 XD, Requires<[HasSSE2]>;
1843 // Convert with truncation packed single/double fp to doubleword
1844 // SSE2 packed instructions with XS prefix
1845 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1848 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1849 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1850 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1851 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1854 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1855 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1856 "cvttps2dq\t{$src, $dst|$dst, $src}",
1858 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1859 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1860 "cvttps2dq\t{$src, $dst|$dst, $src}",
1862 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1864 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1865 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1867 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1868 XS, VEX, Requires<[HasAVX]>;
1869 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1870 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1871 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1872 (memop addr:$src)))]>,
1873 XS, VEX, Requires<[HasAVX]>;
1875 let Predicates = [HasSSE2] in {
1876 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1877 (Int_CVTDQ2PSrr VR128:$src)>;
1878 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1879 (CVTTPS2DQrr VR128:$src)>;
1882 let Predicates = [HasAVX] in {
1883 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1884 (Int_VCVTDQ2PSrr VR128:$src)>;
1885 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1886 (VCVTTPS2DQrr VR128:$src)>;
1887 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1888 (VCVTDQ2PSYrr VR256:$src)>;
1889 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1890 (VCVTTPS2DQYrr VR256:$src)>;
1893 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1894 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1896 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1897 let isCodeGenOnly = 1 in
1898 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1899 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1900 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1901 (memop addr:$src)))]>, VEX;
1902 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1903 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1904 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1905 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1906 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1907 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1908 (memop addr:$src)))]>;
1910 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1911 // register, but the same isn't true when using memory operands instead.
1912 // Provide other assembly rr and rm forms to address this explicitly.
1913 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1914 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1917 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1918 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1919 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1920 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1923 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1924 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1925 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1926 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1928 // Convert packed single to packed double
1929 let Predicates = [HasAVX] in {
1930 // SSE2 instructions without OpSize prefix
1931 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1932 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1933 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1934 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1935 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1936 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1937 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1938 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1940 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1941 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1942 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1943 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1945 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1946 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1947 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1948 TB, VEX, Requires<[HasAVX]>;
1949 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1950 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1951 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1952 (load addr:$src)))]>,
1953 TB, VEX, Requires<[HasAVX]>;
1954 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1955 "cvtps2pd\t{$src, $dst|$dst, $src}",
1956 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1957 TB, Requires<[HasSSE2]>;
1958 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1959 "cvtps2pd\t{$src, $dst|$dst, $src}",
1960 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1961 (load addr:$src)))]>,
1962 TB, Requires<[HasSSE2]>;
1964 // Convert packed double to packed single
1965 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1966 // register, but the same isn't true when using memory operands instead.
1967 // Provide other assembly rr and rm forms to address this explicitly.
1968 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1969 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1970 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1971 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1974 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1975 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1976 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1977 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1980 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1981 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1982 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1983 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1984 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1985 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1986 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1987 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1990 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1991 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1992 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1993 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1995 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1996 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1997 (memop addr:$src)))]>;
1998 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1999 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2000 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2001 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2002 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2003 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2004 (memop addr:$src)))]>;
2006 // AVX 256-bit register conversion intrinsics
2007 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2008 // whenever possible to avoid declaring two versions of each one.
2009 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2010 (VCVTDQ2PSYrr VR256:$src)>;
2011 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2012 (VCVTDQ2PSYrm addr:$src)>;
2014 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2015 (VCVTPD2PSYrr VR256:$src)>;
2016 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2017 (VCVTPD2PSYrm addr:$src)>;
2019 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2020 (VCVTPS2DQYrr VR256:$src)>;
2021 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2022 (VCVTPS2DQYrm addr:$src)>;
2024 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2025 (VCVTPS2PDYrr VR128:$src)>;
2026 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2027 (VCVTPS2PDYrm addr:$src)>;
2029 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2030 (VCVTTPD2DQYrr VR256:$src)>;
2031 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2032 (VCVTTPD2DQYrm addr:$src)>;
2034 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2035 (VCVTTPS2DQYrr VR256:$src)>;
2036 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2037 (VCVTTPS2DQYrm addr:$src)>;
2039 // Match fround and fextend for 128/256-bit conversions
2040 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2041 (VCVTPD2PSYrr VR256:$src)>;
2042 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2043 (VCVTPD2PSYrm addr:$src)>;
2045 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2046 (VCVTPS2PDYrr VR128:$src)>;
2047 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2048 (VCVTPS2PDYrm addr:$src)>;
2050 //===----------------------------------------------------------------------===//
2051 // SSE 1 & 2 - Compare Instructions
2052 //===----------------------------------------------------------------------===//
2054 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2055 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2056 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2057 string asm, string asm_alt> {
2058 def rr : SIi8<0xC2, MRMSrcReg,
2059 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2060 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2061 def rm : SIi8<0xC2, MRMSrcMem,
2062 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2063 [(set RC:$dst, (OpNode (VT RC:$src1),
2064 (ld_frag addr:$src2), imm:$cc))]>;
2066 // Accept explicit immediate argument form instead of comparison code.
2067 let neverHasSideEffects = 1 in {
2068 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2069 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2071 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2072 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2076 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2077 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2078 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2079 XS, VEX_4V, VEX_LIG;
2080 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2081 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2082 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2083 XD, VEX_4V, VEX_LIG;
2085 let Constraints = "$src1 = $dst" in {
2086 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2087 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2088 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2090 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2091 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2092 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2096 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2097 Intrinsic Int, string asm> {
2098 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2099 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2100 [(set VR128:$dst, (Int VR128:$src1,
2101 VR128:$src, imm:$cc))]>;
2102 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2103 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2104 [(set VR128:$dst, (Int VR128:$src1,
2105 (load addr:$src), imm:$cc))]>;
2108 // Aliases to match intrinsics which expect XMM operand(s).
2109 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2110 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2112 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2113 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2115 let Constraints = "$src1 = $dst" in {
2116 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2117 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2118 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2119 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2123 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2124 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2125 ValueType vt, X86MemOperand x86memop,
2126 PatFrag ld_frag, string OpcodeStr, Domain d> {
2127 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2128 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2129 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2130 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2131 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2132 [(set EFLAGS, (OpNode (vt RC:$src1),
2133 (ld_frag addr:$src2)))], d>;
2136 let Defs = [EFLAGS] in {
2137 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2138 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2139 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2140 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2142 let Pattern = []<dag> in {
2143 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2144 "comiss", SSEPackedSingle>, TB, VEX,
2146 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2147 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2151 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2152 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2153 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2154 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2156 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2157 load, "comiss", SSEPackedSingle>, TB, VEX;
2158 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2159 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2160 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2161 "ucomiss", SSEPackedSingle>, TB;
2162 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2163 "ucomisd", SSEPackedDouble>, TB, OpSize;
2165 let Pattern = []<dag> in {
2166 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2167 "comiss", SSEPackedSingle>, TB;
2168 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2169 "comisd", SSEPackedDouble>, TB, OpSize;
2172 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2173 load, "ucomiss", SSEPackedSingle>, TB;
2174 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2175 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2177 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2178 "comiss", SSEPackedSingle>, TB;
2179 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2180 "comisd", SSEPackedDouble>, TB, OpSize;
2181 } // Defs = [EFLAGS]
2183 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2184 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2185 Intrinsic Int, string asm, string asm_alt,
2187 let isAsmParserOnly = 1 in {
2188 def rri : PIi8<0xC2, MRMSrcReg,
2189 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2190 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2191 def rmi : PIi8<0xC2, MRMSrcMem,
2192 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2193 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2196 // Accept explicit immediate argument form instead of comparison code.
2197 def rri_alt : PIi8<0xC2, MRMSrcReg,
2198 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2200 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2201 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2205 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2206 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2207 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2208 SSEPackedSingle>, TB, VEX_4V;
2209 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2210 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2211 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2212 SSEPackedDouble>, TB, OpSize, VEX_4V;
2213 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2214 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2215 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2216 SSEPackedSingle>, TB, VEX_4V;
2217 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2218 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2219 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2220 SSEPackedDouble>, TB, OpSize, VEX_4V;
2221 let Constraints = "$src1 = $dst" in {
2222 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2223 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2224 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2225 SSEPackedSingle>, TB;
2226 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2227 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2228 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2229 SSEPackedDouble>, TB, OpSize;
2232 let Predicates = [HasSSE1] in {
2233 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2234 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2235 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2236 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2239 let Predicates = [HasSSE2] in {
2240 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2241 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2242 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2243 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2246 let Predicates = [HasAVX] in {
2247 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2248 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2249 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2250 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2251 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2252 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2253 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2254 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2256 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2257 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2258 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2259 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2260 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2261 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2262 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2263 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2266 //===----------------------------------------------------------------------===//
2267 // SSE 1 & 2 - Shuffle Instructions
2268 //===----------------------------------------------------------------------===//
2270 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2271 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2272 ValueType vt, string asm, PatFrag mem_frag,
2273 Domain d, bit IsConvertibleToThreeAddress = 0> {
2274 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2275 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2276 [(set RC:$dst, (vt (shufp:$src3
2277 RC:$src1, (mem_frag addr:$src2))))], d>;
2278 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2279 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2280 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2282 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2285 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2286 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2287 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2288 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2289 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2290 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2291 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2292 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2293 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2294 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2295 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2296 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2298 let Constraints = "$src1 = $dst" in {
2299 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2300 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2301 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2303 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2304 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2305 memopv2f64, SSEPackedDouble>, TB, OpSize;
2308 let Predicates = [HasSSE1] in {
2309 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2310 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2311 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2312 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2313 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2314 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2315 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2316 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2317 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2318 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2319 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2320 // fall back to this for SSE1)
2321 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2322 (SHUFPSrri VR128:$src2, VR128:$src1,
2323 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2324 // Special unary SHUFPSrri case.
2325 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2326 (SHUFPSrri VR128:$src1, VR128:$src1,
2327 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 let Predicates = [HasSSE2] in {
2331 // Special binary v4i32 shuffle cases with SHUFPS.
2332 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2333 (SHUFPSrri VR128:$src1, VR128:$src2,
2334 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2335 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2336 (bc_v4i32 (memopv2i64 addr:$src2)))),
2337 (SHUFPSrmi VR128:$src1, addr:$src2,
2338 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2339 // Special unary SHUFPDrri cases.
2340 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2341 (SHUFPDrri VR128:$src1, VR128:$src1,
2342 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2343 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2344 (SHUFPDrri VR128:$src1, VR128:$src1,
2345 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2346 // Special binary v2i64 shuffle cases using SHUFPDrri.
2347 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2348 (SHUFPDrri VR128:$src1, VR128:$src2,
2349 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2350 // Generic SHUFPD patterns
2351 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2352 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2353 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2354 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2355 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2356 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2357 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2358 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2359 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2360 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2363 let Predicates = [HasAVX] in {
2364 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2365 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2366 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2367 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2368 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2369 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2370 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2371 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2372 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2373 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2374 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2375 // fall back to this for SSE1)
2376 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2377 (VSHUFPSrri VR128:$src2, VR128:$src1,
2378 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2379 // Special unary SHUFPSrri case.
2380 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2381 (VSHUFPSrri VR128:$src1, VR128:$src1,
2382 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2383 // Special binary v4i32 shuffle cases with SHUFPS.
2384 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2385 (VSHUFPSrri VR128:$src1, VR128:$src2,
2386 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2387 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2388 (bc_v4i32 (memopv2i64 addr:$src2)))),
2389 (VSHUFPSrmi VR128:$src1, addr:$src2,
2390 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2391 // Special unary SHUFPDrri cases.
2392 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2393 (VSHUFPDrri VR128:$src1, VR128:$src1,
2394 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2395 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2396 (VSHUFPDrri VR128:$src1, VR128:$src1,
2397 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2398 // Special binary v2i64 shuffle cases using SHUFPDrri.
2399 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2400 (VSHUFPDrri VR128:$src1, VR128:$src2,
2401 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2403 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2404 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2405 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2406 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2407 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2408 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2409 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2410 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2411 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2412 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2415 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2416 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2417 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2418 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2419 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2421 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2422 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2423 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2424 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2425 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2427 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2428 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2429 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2430 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2431 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2433 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2434 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2435 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2436 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2437 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2440 //===----------------------------------------------------------------------===//
2441 // SSE 1 & 2 - Unpack Instructions
2442 //===----------------------------------------------------------------------===//
2444 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2445 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2446 PatFrag mem_frag, RegisterClass RC,
2447 X86MemOperand x86memop, string asm,
2449 def rr : PI<opc, MRMSrcReg,
2450 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2452 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2453 def rm : PI<opc, MRMSrcMem,
2454 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2456 (vt (OpNode RC:$src1,
2457 (mem_frag addr:$src2))))], d>;
2460 let AddedComplexity = 10 in {
2461 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2462 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2463 SSEPackedSingle>, TB, VEX_4V;
2464 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2465 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2466 SSEPackedDouble>, TB, OpSize, VEX_4V;
2467 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2468 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 SSEPackedSingle>, TB, VEX_4V;
2470 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2471 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 SSEPackedDouble>, TB, OpSize, VEX_4V;
2474 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2475 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2476 SSEPackedSingle>, TB, VEX_4V;
2477 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2478 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2479 SSEPackedDouble>, TB, OpSize, VEX_4V;
2480 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2481 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2482 SSEPackedSingle>, TB, VEX_4V;
2483 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2484 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2485 SSEPackedDouble>, TB, OpSize, VEX_4V;
2487 let Constraints = "$src1 = $dst" in {
2488 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2489 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2490 SSEPackedSingle>, TB;
2491 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2492 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2493 SSEPackedDouble>, TB, OpSize;
2494 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2495 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2496 SSEPackedSingle>, TB;
2497 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2498 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2499 SSEPackedDouble>, TB, OpSize;
2500 } // Constraints = "$src1 = $dst"
2501 } // AddedComplexity
2503 let Predicates = [HasSSE1] in {
2504 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2505 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2506 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2507 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2508 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2509 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2510 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2511 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2514 let Predicates = [HasSSE2] in {
2515 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2516 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2517 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2518 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2519 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2520 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2521 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2522 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2524 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2525 // problem is during lowering, where it's not possible to recognize the load
2526 // fold cause it has two uses through a bitcast. One use disappears at isel
2527 // time and the fold opportunity reappears.
2528 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2529 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2531 let AddedComplexity = 10 in
2532 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2533 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2536 let Predicates = [HasAVX] in {
2537 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2538 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2539 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2540 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2541 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2542 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2543 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2544 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2546 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2547 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2548 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2549 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2550 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2551 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2552 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2553 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2555 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2556 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2557 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2558 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2559 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2560 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2561 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2562 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2564 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2565 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2566 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2567 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2568 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2569 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2570 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2571 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2573 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2574 // problem is during lowering, where it's not possible to recognize the load
2575 // fold cause it has two uses through a bitcast. One use disappears at isel
2576 // time and the fold opportunity reappears.
2577 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2578 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2579 let AddedComplexity = 10 in
2580 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2581 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2584 //===----------------------------------------------------------------------===//
2585 // SSE 1 & 2 - Extract Floating-Point Sign mask
2586 //===----------------------------------------------------------------------===//
2588 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2589 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2591 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2592 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2593 [(set GR32:$dst, (Int RC:$src))], d>;
2594 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2595 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2598 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2599 SSEPackedSingle>, TB;
2600 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2601 SSEPackedDouble>, TB, OpSize;
2603 def : Pat<(i32 (X86fgetsign FR32:$src)),
2604 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2605 sub_ss))>, Requires<[HasSSE1]>;
2606 def : Pat<(i64 (X86fgetsign FR32:$src)),
2607 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2608 sub_ss))>, Requires<[HasSSE1]>;
2609 def : Pat<(i32 (X86fgetsign FR64:$src)),
2610 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2611 sub_sd))>, Requires<[HasSSE2]>;
2612 def : Pat<(i64 (X86fgetsign FR64:$src)),
2613 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2614 sub_sd))>, Requires<[HasSSE2]>;
2616 let Predicates = [HasAVX] in {
2617 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2618 "movmskps", SSEPackedSingle>, TB, VEX;
2619 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2620 "movmskpd", SSEPackedDouble>, TB,
2622 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2623 "movmskps", SSEPackedSingle>, TB, VEX;
2624 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2625 "movmskpd", SSEPackedDouble>, TB,
2628 def : Pat<(i32 (X86fgetsign FR32:$src)),
2629 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2631 def : Pat<(i64 (X86fgetsign FR32:$src)),
2632 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2634 def : Pat<(i32 (X86fgetsign FR64:$src)),
2635 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2637 def : Pat<(i64 (X86fgetsign FR64:$src)),
2638 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2642 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2643 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2644 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2645 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2647 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2648 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2649 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2650 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2654 //===---------------------------------------------------------------------===//
2655 // SSE2 - Packed Integer Logical Instructions
2656 //===---------------------------------------------------------------------===//
2658 let ExeDomain = SSEPackedInt in { // SSE integer instructions
2660 /// PDI_binop_rm - Simple SSE2 binary operator.
2661 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
2662 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
2663 X86MemOperand x86memop, bit IsCommutable = 0,
2665 let isCommutable = IsCommutable in
2666 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
2667 (ins RC:$src1, RC:$src2),
2669 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2670 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2671 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
2672 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
2673 (ins RC:$src1, x86memop:$src2),
2675 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
2676 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
2677 [(set RC:$dst, (OpVT (OpNode RC:$src1,
2678 (bitconvert (memop_frag addr:$src2)))))]>;
2680 } // ExeDomain = SSEPackedInt
2682 // These are ordered here for pattern ordering requirements with the fp versions
2684 let Predicates = [HasAVX] in {
2685 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
2686 i128mem, 1, 0>, VEX_4V;
2687 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
2688 i128mem, 1, 0>, VEX_4V;
2689 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
2690 i128mem, 1, 0>, VEX_4V;
2691 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
2692 i128mem, 0, 0>, VEX_4V;
2695 let Constraints = "$src1 = $dst" in {
2696 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
2698 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
2700 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
2702 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
2704 } // Constraints = "$src1 = $dst"
2706 let Predicates = [HasAVX2] in {
2707 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
2708 i256mem, 1, 0>, VEX_4V;
2709 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
2710 i256mem, 1, 0>, VEX_4V;
2711 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
2712 i256mem, 1, 0>, VEX_4V;
2713 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
2714 i256mem, 0, 0>, VEX_4V;
2717 //===----------------------------------------------------------------------===//
2718 // SSE 1 & 2 - Logical Instructions
2719 //===----------------------------------------------------------------------===//
2721 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2723 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2725 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2726 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2728 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2729 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2731 let Constraints = "$src1 = $dst" in {
2732 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2733 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2735 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2736 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2740 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2741 let mayLoad = 0 in {
2742 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2743 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2744 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2747 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2748 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2750 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2752 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2754 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2755 // are all promoted to v2i64, and the patterns are covered by the int
2756 // version. This is needed in SSE only, because v2i64 isn't supported on
2757 // SSE1, but only on SSE2.
2758 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2759 !strconcat(OpcodeStr, "ps"), f128mem, [],
2760 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2761 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2763 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2764 !strconcat(OpcodeStr, "pd"), f128mem,
2765 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2766 (bc_v2i64 (v2f64 VR128:$src2))))],
2767 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2768 (memopv2i64 addr:$src2)))], 0>,
2770 let Constraints = "$src1 = $dst" in {
2771 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2772 !strconcat(OpcodeStr, "ps"), f128mem,
2773 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2774 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2775 (memopv2i64 addr:$src2)))]>, TB;
2777 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2778 !strconcat(OpcodeStr, "pd"), f128mem,
2779 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2780 (bc_v2i64 (v2f64 VR128:$src2))))],
2781 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2782 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2786 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2788 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2790 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2791 !strconcat(OpcodeStr, "ps"), f256mem,
2792 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2793 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2794 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2796 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2797 !strconcat(OpcodeStr, "pd"), f256mem,
2798 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2799 (bc_v4i64 (v4f64 VR256:$src2))))],
2800 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2801 (memopv4i64 addr:$src2)))], 0>,
2805 // AVX 256-bit packed logical ops forms
2806 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2807 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2808 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2809 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2811 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2812 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2813 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2814 let isCommutable = 0 in
2815 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2817 //===----------------------------------------------------------------------===//
2818 // SSE 1 & 2 - Arithmetic Instructions
2819 //===----------------------------------------------------------------------===//
2821 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2824 /// In addition, we also have a special variant of the scalar form here to
2825 /// represent the associated intrinsic operation. This form is unlike the
2826 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2827 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2829 /// These three forms can each be reg+reg or reg+mem.
2832 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2834 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2836 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2837 OpNode, FR32, f32mem, Is2Addr>, XS;
2838 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2839 OpNode, FR64, f64mem, Is2Addr>, XD;
2842 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2844 let mayLoad = 0 in {
2845 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2846 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2847 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2848 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2852 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2854 let mayLoad = 0 in {
2855 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2856 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2857 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2858 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2862 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2864 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2865 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2866 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2867 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2870 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2872 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2873 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2874 SSEPackedSingle, Is2Addr>, TB;
2876 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2877 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2878 SSEPackedDouble, Is2Addr>, TB, OpSize;
2881 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2882 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2883 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2884 SSEPackedSingle, 0>, TB;
2886 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2887 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2888 SSEPackedDouble, 0>, TB, OpSize;
2891 // Binary Arithmetic instructions
2892 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2893 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2894 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2895 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2896 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2897 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2898 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2899 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2901 let isCommutable = 0 in {
2902 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2903 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2904 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2905 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2906 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2907 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2908 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2909 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2910 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2911 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2912 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2913 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2914 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2915 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2916 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2917 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2918 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2919 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2920 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2921 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2924 let Constraints = "$src1 = $dst" in {
2925 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2926 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2927 basic_sse12_fp_binop_s_int<0x58, "add">;
2928 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2929 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2930 basic_sse12_fp_binop_s_int<0x59, "mul">;
2932 let isCommutable = 0 in {
2933 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2934 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2935 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2936 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2937 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2938 basic_sse12_fp_binop_s_int<0x5E, "div">;
2939 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2940 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2941 basic_sse12_fp_binop_s_int<0x5F, "max">,
2942 basic_sse12_fp_binop_p_int<0x5F, "max">;
2943 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2944 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2945 basic_sse12_fp_binop_s_int<0x5D, "min">,
2946 basic_sse12_fp_binop_p_int<0x5D, "min">;
2951 /// In addition, we also have a special variant of the scalar form here to
2952 /// represent the associated intrinsic operation. This form is unlike the
2953 /// plain scalar form, in that it takes an entire vector (instead of a
2954 /// scalar) and leaves the top elements undefined.
2956 /// And, we have a special variant form for a full-vector intrinsic form.
2958 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2959 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2960 SDNode OpNode, Intrinsic F32Int> {
2961 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2962 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2963 [(set FR32:$dst, (OpNode FR32:$src))]>;
2964 // For scalar unary operations, fold a load into the operation
2965 // only in OptForSize mode. It eliminates an instruction, but it also
2966 // eliminates a whole-register clobber (the load), so it introduces a
2967 // partial register update condition.
2968 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2969 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2970 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2971 Requires<[HasSSE1, OptForSize]>;
2972 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2973 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2974 [(set VR128:$dst, (F32Int VR128:$src))]>;
2975 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2976 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2977 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2980 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2981 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2982 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2983 !strconcat(OpcodeStr,
2984 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2986 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2987 !strconcat(OpcodeStr,
2988 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2989 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2990 (ins VR128:$src1, ssmem:$src2),
2991 !strconcat(OpcodeStr,
2992 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2995 /// sse1_fp_unop_p - SSE1 unops in packed form.
2996 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2997 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2998 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2999 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
3000 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3001 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3002 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
3005 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
3006 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3007 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3008 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3009 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
3010 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3011 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3012 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
3015 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
3016 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3017 Intrinsic V4F32Int> {
3018 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3019 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3020 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
3021 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3022 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3023 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
3026 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
3027 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3028 Intrinsic V4F32Int> {
3029 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3030 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3031 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
3032 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3033 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
3034 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
3037 /// sse2_fp_unop_s - SSE2 unops in scalar form.
3038 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
3039 SDNode OpNode, Intrinsic F64Int> {
3040 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
3041 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3042 [(set FR64:$dst, (OpNode FR64:$src))]>;
3043 // See the comments in sse1_fp_unop_s for why this is OptForSize.
3044 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
3045 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3046 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
3047 Requires<[HasSSE2, OptForSize]>;
3048 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3049 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3050 [(set VR128:$dst, (F64Int VR128:$src))]>;
3051 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
3052 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
3053 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
3056 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
3057 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
3058 let neverHasSideEffects = 1 in {
3059 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
3060 !strconcat(OpcodeStr,
3061 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3063 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3064 !strconcat(OpcodeStr,
3065 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3067 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3068 (ins VR128:$src1, sdmem:$src2),
3069 !strconcat(OpcodeStr,
3070 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3073 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3074 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3076 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3077 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3078 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3079 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3080 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3081 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3084 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3085 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3086 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3087 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3088 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3089 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3090 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3091 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3094 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3095 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3096 Intrinsic V2F64Int> {
3097 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3098 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3099 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3100 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3101 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3102 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3105 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3106 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3107 Intrinsic V2F64Int> {
3108 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3109 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3110 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3111 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3112 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3113 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3116 let Predicates = [HasAVX] in {
3118 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3119 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3121 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3122 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3123 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3124 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3125 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3126 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3127 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3128 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3131 // Reciprocal approximations. Note that these typically require refinement
3132 // in order to obtain suitable precision.
3133 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3134 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3135 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3136 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3137 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3139 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3140 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3141 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3142 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3143 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3146 def : Pat<(f32 (fsqrt FR32:$src)),
3147 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3148 def : Pat<(f32 (fsqrt (load addr:$src))),
3149 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3150 Requires<[HasAVX, OptForSize]>;
3151 def : Pat<(f64 (fsqrt FR64:$src)),
3152 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3153 def : Pat<(f64 (fsqrt (load addr:$src))),
3154 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3155 Requires<[HasAVX, OptForSize]>;
3157 def : Pat<(f32 (X86frsqrt FR32:$src)),
3158 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3159 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3160 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3161 Requires<[HasAVX, OptForSize]>;
3163 def : Pat<(f32 (X86frcp FR32:$src)),
3164 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3165 def : Pat<(f32 (X86frcp (load addr:$src))),
3166 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3167 Requires<[HasAVX, OptForSize]>;
3169 let Predicates = [HasAVX] in {
3170 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3171 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3172 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3173 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3175 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3176 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3178 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3179 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3180 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3181 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3183 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3184 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3186 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3187 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3188 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3189 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3191 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3192 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3194 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3195 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3196 (VRCPSSr (f32 (IMPLICIT_DEF)),
3197 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3199 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3200 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3204 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3205 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3206 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3207 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3208 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3209 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3211 // Reciprocal approximations. Note that these typically require refinement
3212 // in order to obtain suitable precision.
3213 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3214 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3215 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3216 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3217 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3218 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3220 // There is no f64 version of the reciprocal approximation instructions.
3222 //===----------------------------------------------------------------------===//
3223 // SSE 1 & 2 - Non-temporal stores
3224 //===----------------------------------------------------------------------===//
3226 let AddedComplexity = 400 in { // Prefer non-temporal versions
3227 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3228 (ins f128mem:$dst, VR128:$src),
3229 "movntps\t{$src, $dst|$dst, $src}",
3230 [(alignednontemporalstore (v4f32 VR128:$src),
3232 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3233 (ins f128mem:$dst, VR128:$src),
3234 "movntpd\t{$src, $dst|$dst, $src}",
3235 [(alignednontemporalstore (v2f64 VR128:$src),
3237 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3238 (ins f128mem:$dst, VR128:$src),
3239 "movntdq\t{$src, $dst|$dst, $src}",
3240 [(alignednontemporalstore (v2f64 VR128:$src),
3243 let ExeDomain = SSEPackedInt in
3244 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3245 (ins f128mem:$dst, VR128:$src),
3246 "movntdq\t{$src, $dst|$dst, $src}",
3247 [(alignednontemporalstore (v4f32 VR128:$src),
3250 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3251 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3253 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3254 (ins f256mem:$dst, VR256:$src),
3255 "movntps\t{$src, $dst|$dst, $src}",
3256 [(alignednontemporalstore (v8f32 VR256:$src),
3258 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3259 (ins f256mem:$dst, VR256:$src),
3260 "movntpd\t{$src, $dst|$dst, $src}",
3261 [(alignednontemporalstore (v4f64 VR256:$src),
3263 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3264 (ins f256mem:$dst, VR256:$src),
3265 "movntdq\t{$src, $dst|$dst, $src}",
3266 [(alignednontemporalstore (v4f64 VR256:$src),
3268 let ExeDomain = SSEPackedInt in
3269 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3270 (ins f256mem:$dst, VR256:$src),
3271 "movntdq\t{$src, $dst|$dst, $src}",
3272 [(alignednontemporalstore (v8f32 VR256:$src),
3276 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3277 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3278 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3279 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3280 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3281 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3283 let AddedComplexity = 400 in { // Prefer non-temporal versions
3284 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3285 "movntps\t{$src, $dst|$dst, $src}",
3286 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3287 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3288 "movntpd\t{$src, $dst|$dst, $src}",
3289 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3291 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3292 "movntdq\t{$src, $dst|$dst, $src}",
3293 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3295 let ExeDomain = SSEPackedInt in
3296 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3297 "movntdq\t{$src, $dst|$dst, $src}",
3298 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3300 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3301 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3303 // There is no AVX form for instructions below this point
3304 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3305 "movnti{l}\t{$src, $dst|$dst, $src}",
3306 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3307 TB, Requires<[HasXMMInt]>;
3308 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3309 "movnti{q}\t{$src, $dst|$dst, $src}",
3310 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3311 TB, Requires<[HasXMMInt]>;
3314 //===----------------------------------------------------------------------===//
3315 // SSE 1 & 2 - Prefetch and memory fence
3316 //===----------------------------------------------------------------------===//
3318 // Prefetch intrinsic.
3319 let Predicates = [HasXMM] in {
3320 def PREFETCHT0 : I<0x18, MRM1m, (outs), (ins i8mem:$src),
3321 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>, TB;
3322 def PREFETCHT1 : I<0x18, MRM2m, (outs), (ins i8mem:$src),
3323 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>, TB;
3324 def PREFETCHT2 : I<0x18, MRM3m, (outs), (ins i8mem:$src),
3325 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>, TB;
3326 def PREFETCHNTA : I<0x18, MRM0m, (outs), (ins i8mem:$src),
3327 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>, TB;
3331 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3332 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3333 TB, Requires<[HasXMMInt]>;
3335 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3336 // was introduced with SSE2, it's backward compatible.
3337 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3339 // Load, store, and memory fence
3340 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3341 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasXMM]>;
3342 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3343 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasXMMInt]>;
3344 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3345 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasXMMInt]>;
3347 def : Pat<(X86SFence), (SFENCE)>;
3348 def : Pat<(X86LFence), (LFENCE)>;
3349 def : Pat<(X86MFence), (MFENCE)>;
3351 //===----------------------------------------------------------------------===//
3352 // SSE 1 & 2 - Load/Store XCSR register
3353 //===----------------------------------------------------------------------===//
3355 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3356 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3357 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3358 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3360 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3361 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3362 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3363 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3365 //===---------------------------------------------------------------------===//
3366 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3367 //===---------------------------------------------------------------------===//
3369 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3371 let neverHasSideEffects = 1 in {
3372 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3373 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3374 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3375 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3377 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3378 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3379 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3380 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3383 let isCodeGenOnly = 1 in {
3384 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3385 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3386 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3387 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3388 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3389 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3390 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3391 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3394 let canFoldAsLoad = 1, mayLoad = 1 in {
3395 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3396 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3397 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3398 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3399 let Predicates = [HasAVX] in {
3400 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3401 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3402 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3403 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3407 let mayStore = 1 in {
3408 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3409 (ins i128mem:$dst, VR128:$src),
3410 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3411 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3412 (ins i256mem:$dst, VR256:$src),
3413 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3414 let Predicates = [HasAVX] in {
3415 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3416 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3417 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3418 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3422 let neverHasSideEffects = 1 in
3423 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3424 "movdqa\t{$src, $dst|$dst, $src}", []>;
3426 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3427 "movdqu\t{$src, $dst|$dst, $src}",
3428 []>, XS, Requires<[HasSSE2]>;
3431 let isCodeGenOnly = 1 in {
3432 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3433 "movdqa\t{$src, $dst|$dst, $src}", []>;
3435 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3436 "movdqu\t{$src, $dst|$dst, $src}",
3437 []>, XS, Requires<[HasSSE2]>;
3440 let canFoldAsLoad = 1, mayLoad = 1 in {
3441 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3442 "movdqa\t{$src, $dst|$dst, $src}",
3443 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3444 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3445 "movdqu\t{$src, $dst|$dst, $src}",
3446 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3447 XS, Requires<[HasSSE2]>;
3450 let mayStore = 1 in {
3451 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3452 "movdqa\t{$src, $dst|$dst, $src}",
3453 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3454 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3455 "movdqu\t{$src, $dst|$dst, $src}",
3456 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3457 XS, Requires<[HasSSE2]>;
3460 // Intrinsic forms of MOVDQU load and store
3461 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3462 "vmovdqu\t{$src, $dst|$dst, $src}",
3463 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3464 XS, VEX, Requires<[HasAVX]>;
3466 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3467 "movdqu\t{$src, $dst|$dst, $src}",
3468 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3469 XS, Requires<[HasSSE2]>;
3471 } // ExeDomain = SSEPackedInt
3473 let Predicates = [HasAVX] in {
3474 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3475 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3476 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3479 //===---------------------------------------------------------------------===//
3480 // SSE2 - Packed Integer Arithmetic Instructions
3481 //===---------------------------------------------------------------------===//
3483 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3485 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3486 RegisterClass RC, PatFrag memop_frag,
3487 X86MemOperand x86memop, bit IsCommutable = 0,
3489 let isCommutable = IsCommutable in
3490 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3491 (ins RC:$src1, RC:$src2),
3493 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3494 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3495 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3496 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3497 (ins RC:$src1, x86memop:$src2),
3499 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3500 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3501 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3504 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3505 string OpcodeStr, Intrinsic IntId,
3506 Intrinsic IntId2, RegisterClass RC,
3508 // src2 is always 128-bit
3509 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3510 (ins RC:$src1, VR128:$src2),
3512 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3513 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3514 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3515 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3516 (ins RC:$src1, i128mem:$src2),
3518 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3519 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3520 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3521 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3522 (ins RC:$src1, i32i8imm:$src2),
3524 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3525 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3526 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3529 } // ExeDomain = SSEPackedInt
3531 // 128-bit Integer Arithmetic
3533 let Predicates = [HasAVX] in {
3534 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3535 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3536 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3537 i128mem, 1, 0>, VEX_4V;
3538 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3539 i128mem, 1, 0>, VEX_4V;
3540 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3541 i128mem, 1, 0>, VEX_4V;
3542 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3543 i128mem, 1, 0>, VEX_4V;
3544 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3545 i128mem, 0, 0>, VEX_4V;
3546 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3547 i128mem, 0, 0>, VEX_4V;
3548 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3549 i128mem, 0, 0>, VEX_4V;
3550 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3551 i128mem, 0, 0>, VEX_4V;
3554 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3555 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3556 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3557 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3558 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3559 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3560 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3561 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3562 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3563 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3564 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3565 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3566 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3567 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3568 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3569 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3570 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3571 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3572 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3573 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3574 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3575 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3576 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3577 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3578 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3579 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3580 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3581 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3582 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3583 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3584 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3585 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3586 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3587 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3588 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3589 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3590 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3591 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3594 let Predicates = [HasAVX2] in {
3595 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3596 i256mem, 1, 0>, VEX_4V;
3597 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3598 i256mem, 1, 0>, VEX_4V;
3599 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3600 i256mem, 1, 0>, VEX_4V;
3601 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3602 i256mem, 1, 0>, VEX_4V;
3603 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3604 i256mem, 1, 0>, VEX_4V;
3605 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3606 i256mem, 0, 0>, VEX_4V;
3607 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3608 i256mem, 0, 0>, VEX_4V;
3609 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3610 i256mem, 0, 0>, VEX_4V;
3611 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3612 i256mem, 0, 0>, VEX_4V;
3615 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3616 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3617 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3618 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3619 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3620 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3621 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3622 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3623 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3624 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3625 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3626 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3627 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3628 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3629 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3630 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3631 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3632 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3633 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3634 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3635 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3636 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3637 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3638 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3639 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3640 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3641 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3642 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3643 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3644 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3645 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3646 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3647 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3648 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3649 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3650 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3651 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3652 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3655 let Constraints = "$src1 = $dst" in {
3656 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3658 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3660 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3662 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3664 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3666 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3668 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3670 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3672 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3676 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3677 VR128, memopv2i64, i128mem>;
3678 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3679 VR128, memopv2i64, i128mem>;
3680 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3681 VR128, memopv2i64, i128mem>;
3682 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3683 VR128, memopv2i64, i128mem>;
3684 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3685 VR128, memopv2i64, i128mem, 1>;
3686 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3687 VR128, memopv2i64, i128mem, 1>;
3688 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3689 VR128, memopv2i64, i128mem, 1>;
3690 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3691 VR128, memopv2i64, i128mem, 1>;
3692 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3693 VR128, memopv2i64, i128mem, 1>;
3694 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3695 VR128, memopv2i64, i128mem, 1>;
3696 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3697 VR128, memopv2i64, i128mem, 1>;
3698 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3699 VR128, memopv2i64, i128mem, 1>;
3700 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3701 VR128, memopv2i64, i128mem, 1>;
3702 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3703 VR128, memopv2i64, i128mem, 1>;
3704 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3705 VR128, memopv2i64, i128mem, 1>;
3706 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3707 VR128, memopv2i64, i128mem, 1>;
3708 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3709 VR128, memopv2i64, i128mem, 1>;
3710 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3711 VR128, memopv2i64, i128mem, 1>;
3712 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3713 VR128, memopv2i64, i128mem, 1>;
3715 } // Constraints = "$src1 = $dst"
3717 //===---------------------------------------------------------------------===//
3718 // SSE2 - Packed Integer Logical Instructions
3719 //===---------------------------------------------------------------------===//
3721 let Predicates = [HasAVX] in {
3722 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3723 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3725 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3726 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3728 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3729 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3732 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3733 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3735 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3736 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3738 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3739 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3742 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3743 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3745 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3746 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3749 let ExeDomain = SSEPackedInt in {
3750 let neverHasSideEffects = 1 in {
3751 // 128-bit logical shifts.
3752 def VPSLLDQri : PDIi8<0x73, MRM7r,
3753 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3754 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3756 def VPSRLDQri : PDIi8<0x73, MRM3r,
3757 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3758 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3760 // PSRADQri doesn't exist in SSE[1-3].
3765 let Predicates = [HasAVX2] in {
3766 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3767 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3769 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3770 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3772 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3773 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3776 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3777 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3779 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3780 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3782 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3783 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3786 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3787 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3789 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3790 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3793 let ExeDomain = SSEPackedInt in {
3794 let neverHasSideEffects = 1 in {
3795 // 128-bit logical shifts.
3796 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3797 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3798 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3800 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3801 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3802 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3804 // PSRADQYri doesn't exist in SSE[1-3].
3809 let Constraints = "$src1 = $dst" in {
3810 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3811 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3813 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3814 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3816 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3817 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3820 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3821 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3823 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3824 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3826 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3827 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3830 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3831 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3833 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3834 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3837 let ExeDomain = SSEPackedInt in {
3838 let neverHasSideEffects = 1 in {
3839 // 128-bit logical shifts.
3840 def PSLLDQri : PDIi8<0x73, MRM7r,
3841 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3842 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3843 def PSRLDQri : PDIi8<0x73, MRM3r,
3844 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3845 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3846 // PSRADQri doesn't exist in SSE[1-3].
3849 } // Constraints = "$src1 = $dst"
3851 let Predicates = [HasAVX] in {
3852 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3853 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3854 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3855 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3856 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3857 (VPSLLDQri VR128:$src1, imm:$src2)>;
3858 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3859 (VPSRLDQri VR128:$src1, imm:$src2)>;
3860 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3861 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3863 // Shift up / down and insert zero's.
3864 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3865 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3866 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3867 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3870 let Predicates = [HasAVX2] in {
3871 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3872 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3873 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3874 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3875 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3876 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3877 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3878 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3881 let Predicates = [HasSSE2] in {
3882 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3883 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3884 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3885 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3886 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3887 (PSLLDQri VR128:$src1, imm:$src2)>;
3888 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3889 (PSRLDQri VR128:$src1, imm:$src2)>;
3890 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3891 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3893 // Shift up / down and insert zero's.
3894 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3895 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3896 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3897 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3900 //===---------------------------------------------------------------------===//
3901 // SSE2 - Packed Integer Comparison Instructions
3902 //===---------------------------------------------------------------------===//
3904 let Predicates = [HasAVX] in {
3905 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3906 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3907 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3908 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3909 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3910 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3911 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3912 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3913 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3914 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3915 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3916 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3918 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3919 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3920 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3921 (bc_v16i8 (memopv2i64 addr:$src2)))),
3922 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3923 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3924 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3925 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3926 (bc_v8i16 (memopv2i64 addr:$src2)))),
3927 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3928 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3929 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3930 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3931 (bc_v4i32 (memopv2i64 addr:$src2)))),
3932 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3934 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3935 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3936 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3937 (bc_v16i8 (memopv2i64 addr:$src2)))),
3938 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3939 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3940 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3941 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3942 (bc_v8i16 (memopv2i64 addr:$src2)))),
3943 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3944 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3945 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3946 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3947 (bc_v4i32 (memopv2i64 addr:$src2)))),
3948 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3951 let Predicates = [HasAVX2] in {
3952 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3953 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3954 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3955 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3956 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3957 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3958 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3959 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3960 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3961 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3962 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3963 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3965 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3966 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3967 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3968 (bc_v32i8 (memopv4i64 addr:$src2)))),
3969 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3970 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3971 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3972 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3973 (bc_v16i16 (memopv4i64 addr:$src2)))),
3974 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3975 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3976 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3977 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3978 (bc_v8i32 (memopv4i64 addr:$src2)))),
3979 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3981 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3982 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3983 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3984 (bc_v32i8 (memopv4i64 addr:$src2)))),
3985 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3986 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3987 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3988 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3989 (bc_v16i16 (memopv4i64 addr:$src2)))),
3990 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3991 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3992 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3993 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3994 (bc_v8i32 (memopv4i64 addr:$src2)))),
3995 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3998 let Constraints = "$src1 = $dst" in {
3999 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
4000 VR128, memopv2i64, i128mem, 1>;
4001 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
4002 VR128, memopv2i64, i128mem, 1>;
4003 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
4004 VR128, memopv2i64, i128mem, 1>;
4005 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
4006 VR128, memopv2i64, i128mem>;
4007 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
4008 VR128, memopv2i64, i128mem>;
4009 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
4010 VR128, memopv2i64, i128mem>;
4011 } // Constraints = "$src1 = $dst"
4013 let Predicates = [HasSSE2] in {
4014 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
4015 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
4016 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
4017 (bc_v16i8 (memopv2i64 addr:$src2)))),
4018 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4019 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4020 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4021 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4022 (bc_v8i16 (memopv2i64 addr:$src2)))),
4023 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4024 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4025 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4026 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4027 (bc_v4i32 (memopv2i64 addr:$src2)))),
4028 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4030 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4031 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4032 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4033 (bc_v16i8 (memopv2i64 addr:$src2)))),
4034 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4035 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4036 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4037 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4038 (bc_v8i16 (memopv2i64 addr:$src2)))),
4039 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4040 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4041 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4042 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4043 (bc_v4i32 (memopv2i64 addr:$src2)))),
4044 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4047 //===---------------------------------------------------------------------===//
4048 // SSE2 - Packed Integer Pack Instructions
4049 //===---------------------------------------------------------------------===//
4051 let Predicates = [HasAVX] in {
4052 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4053 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4054 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4055 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4056 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4057 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4060 let Predicates = [HasAVX2] in {
4061 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4062 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4063 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4064 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4065 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4066 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4069 let Constraints = "$src1 = $dst" in {
4070 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4071 VR128, memopv2i64, i128mem>;
4072 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4073 VR128, memopv2i64, i128mem>;
4074 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4075 VR128, memopv2i64, i128mem>;
4076 } // Constraints = "$src1 = $dst"
4078 //===---------------------------------------------------------------------===//
4079 // SSE2 - Packed Integer Shuffle Instructions
4080 //===---------------------------------------------------------------------===//
4082 let ExeDomain = SSEPackedInt in {
4083 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4085 def ri : Ii8<0x70, MRMSrcReg,
4086 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4087 !strconcat(OpcodeStr,
4088 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4089 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4091 def mi : Ii8<0x70, MRMSrcMem,
4092 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4093 !strconcat(OpcodeStr,
4094 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4095 [(set VR128:$dst, (vt (pshuf_frag:$src2
4096 (bc_frag (memopv2i64 addr:$src1)),
4100 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4102 def Yri : Ii8<0x70, MRMSrcReg,
4103 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4104 !strconcat(OpcodeStr,
4105 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4106 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4108 def Ymi : Ii8<0x70, MRMSrcMem,
4109 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4110 !strconcat(OpcodeStr,
4111 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4112 [(set VR256:$dst, (vt (pshuf_frag:$src2
4113 (bc_frag (memopv4i64 addr:$src1)),
4116 } // ExeDomain = SSEPackedInt
4118 let Predicates = [HasAVX] in {
4119 let AddedComplexity = 5 in
4120 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4123 // SSE2 with ImmT == Imm8 and XS prefix.
4124 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4127 // SSE2 with ImmT == Imm8 and XD prefix.
4128 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4131 let AddedComplexity = 5 in
4132 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4133 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4134 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4135 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4136 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4138 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4140 (VPSHUFDmi addr:$src1, imm:$imm)>;
4141 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4143 (VPSHUFDmi addr:$src1, imm:$imm)>;
4144 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4145 (VPSHUFDri VR128:$src1, imm:$imm)>;
4146 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4147 (VPSHUFDri VR128:$src1, imm:$imm)>;
4148 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4149 (VPSHUFHWri VR128:$src, imm:$imm)>;
4150 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4152 (VPSHUFHWmi addr:$src, imm:$imm)>;
4153 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4154 (VPSHUFLWri VR128:$src, imm:$imm)>;
4155 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4157 (VPSHUFLWmi addr:$src, imm:$imm)>;
4160 let Predicates = [HasAVX2] in {
4161 let AddedComplexity = 5 in
4162 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4165 // SSE2 with ImmT == Imm8 and XS prefix.
4166 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4169 // SSE2 with ImmT == Imm8 and XD prefix.
4170 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4174 let Predicates = [HasSSE2] in {
4175 let AddedComplexity = 5 in
4176 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4178 // SSE2 with ImmT == Imm8 and XS prefix.
4179 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4181 // SSE2 with ImmT == Imm8 and XD prefix.
4182 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4184 let AddedComplexity = 5 in
4185 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4186 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4187 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4188 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4189 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4191 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4193 (PSHUFDmi addr:$src1, imm:$imm)>;
4194 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4196 (PSHUFDmi addr:$src1, imm:$imm)>;
4197 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4198 (PSHUFDri VR128:$src1, imm:$imm)>;
4199 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4200 (PSHUFDri VR128:$src1, imm:$imm)>;
4201 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4202 (PSHUFHWri VR128:$src, imm:$imm)>;
4203 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4205 (PSHUFHWmi addr:$src, imm:$imm)>;
4206 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4207 (PSHUFLWri VR128:$src, imm:$imm)>;
4208 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4210 (PSHUFLWmi addr:$src, imm:$imm)>;
4213 //===---------------------------------------------------------------------===//
4214 // SSE2 - Packed Integer Unpack Instructions
4215 //===---------------------------------------------------------------------===//
4217 let ExeDomain = SSEPackedInt in {
4218 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4219 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4220 def rr : PDI<opc, MRMSrcReg,
4221 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4223 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4224 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4225 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4226 def rm : PDI<opc, MRMSrcMem,
4227 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4229 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4230 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4231 [(set VR128:$dst, (OpNode VR128:$src1,
4232 (bc_frag (memopv2i64
4236 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4237 SDNode OpNode, PatFrag bc_frag> {
4238 def Yrr : PDI<opc, MRMSrcReg,
4239 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4240 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4241 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4242 def Yrm : PDI<opc, MRMSrcMem,
4243 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4244 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4245 [(set VR256:$dst, (OpNode VR256:$src1,
4246 (bc_frag (memopv4i64 addr:$src2))))]>;
4249 let Predicates = [HasAVX] in {
4250 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4251 bc_v16i8, 0>, VEX_4V;
4252 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4253 bc_v8i16, 0>, VEX_4V;
4254 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4255 bc_v4i32, 0>, VEX_4V;
4256 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4257 bc_v2i64, 0>, VEX_4V;
4259 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4260 bc_v16i8, 0>, VEX_4V;
4261 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4262 bc_v8i16, 0>, VEX_4V;
4263 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4264 bc_v4i32, 0>, VEX_4V;
4265 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4266 bc_v2i64, 0>, VEX_4V;
4269 let Predicates = [HasAVX2] in {
4270 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4272 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4274 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4276 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4279 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4281 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4283 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4285 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4289 let Constraints = "$src1 = $dst" in {
4290 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4292 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4294 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4296 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4299 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4301 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4303 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4305 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4308 } // ExeDomain = SSEPackedInt
4310 // Patterns for using AVX1 instructions with integer vectors
4311 // Here to give AVX2 priority
4312 let Predicates = [HasAVX] in {
4313 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4314 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4315 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4316 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4317 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4318 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4319 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4320 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4322 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4323 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4324 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4325 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4326 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4327 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4328 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4329 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4332 // Splat v2f64 / v2i64
4333 let AddedComplexity = 10 in {
4334 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4335 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4336 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4337 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4340 //===---------------------------------------------------------------------===//
4341 // SSE2 - Packed Integer Extract and Insert
4342 //===---------------------------------------------------------------------===//
4344 let ExeDomain = SSEPackedInt in {
4345 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4346 def rri : Ii8<0xC4, MRMSrcReg,
4347 (outs VR128:$dst), (ins VR128:$src1,
4348 GR32:$src2, i32i8imm:$src3),
4350 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4351 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4353 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4354 def rmi : Ii8<0xC4, MRMSrcMem,
4355 (outs VR128:$dst), (ins VR128:$src1,
4356 i16mem:$src2, i32i8imm:$src3),
4358 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4359 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4361 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4366 let Predicates = [HasAVX] in
4367 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4368 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4369 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4370 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4371 imm:$src2))]>, TB, OpSize, VEX;
4372 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4373 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4374 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4375 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4379 let Predicates = [HasAVX] in {
4380 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4381 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4382 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4383 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4384 []>, TB, OpSize, VEX_4V;
4387 let Constraints = "$src1 = $dst" in
4388 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4390 } // ExeDomain = SSEPackedInt
4392 //===---------------------------------------------------------------------===//
4393 // SSE2 - Packed Mask Creation
4394 //===---------------------------------------------------------------------===//
4396 let ExeDomain = SSEPackedInt in {
4398 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4399 "pmovmskb\t{$src, $dst|$dst, $src}",
4400 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4401 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4402 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4404 let Predicates = [HasAVX2] in {
4405 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4406 "pmovmskb\t{$src, $dst|$dst, $src}",
4407 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4408 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4409 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4412 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4413 "pmovmskb\t{$src, $dst|$dst, $src}",
4414 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4416 } // ExeDomain = SSEPackedInt
4418 //===---------------------------------------------------------------------===//
4419 // SSE2 - Conditional Store
4420 //===---------------------------------------------------------------------===//
4422 let ExeDomain = SSEPackedInt in {
4425 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4426 (ins VR128:$src, VR128:$mask),
4427 "maskmovdqu\t{$mask, $src|$src, $mask}",
4428 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4430 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4431 (ins VR128:$src, VR128:$mask),
4432 "maskmovdqu\t{$mask, $src|$src, $mask}",
4433 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4436 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4437 "maskmovdqu\t{$mask, $src|$src, $mask}",
4438 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4440 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4441 "maskmovdqu\t{$mask, $src|$src, $mask}",
4442 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4444 } // ExeDomain = SSEPackedInt
4446 //===---------------------------------------------------------------------===//
4447 // SSE2 - Move Doubleword
4448 //===---------------------------------------------------------------------===//
4450 //===---------------------------------------------------------------------===//
4451 // Move Int Doubleword to Packed Double Int
4453 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4454 "movd\t{$src, $dst|$dst, $src}",
4456 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4457 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4460 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4462 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4463 "mov{d|q}\t{$src, $dst|$dst, $src}",
4465 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4466 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4467 "mov{d|q}\t{$src, $dst|$dst, $src}",
4468 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4470 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4471 "movd\t{$src, $dst|$dst, $src}",
4473 (v4i32 (scalar_to_vector GR32:$src)))]>;
4474 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4475 "movd\t{$src, $dst|$dst, $src}",
4477 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4478 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4479 "mov{d|q}\t{$src, $dst|$dst, $src}",
4481 (v2i64 (scalar_to_vector GR64:$src)))]>;
4482 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4483 "mov{d|q}\t{$src, $dst|$dst, $src}",
4484 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4486 //===---------------------------------------------------------------------===//
4487 // Move Int Doubleword to Single Scalar
4489 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4490 "movd\t{$src, $dst|$dst, $src}",
4491 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4493 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4494 "movd\t{$src, $dst|$dst, $src}",
4495 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4497 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4498 "movd\t{$src, $dst|$dst, $src}",
4499 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4501 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4502 "movd\t{$src, $dst|$dst, $src}",
4503 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4505 //===---------------------------------------------------------------------===//
4506 // Move Packed Doubleword Int to Packed Double Int
4508 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4509 "movd\t{$src, $dst|$dst, $src}",
4510 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4512 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4513 (ins i32mem:$dst, VR128:$src),
4514 "movd\t{$src, $dst|$dst, $src}",
4515 [(store (i32 (vector_extract (v4i32 VR128:$src),
4516 (iPTR 0))), addr:$dst)]>, VEX;
4517 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4518 "movd\t{$src, $dst|$dst, $src}",
4519 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4521 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4522 "movd\t{$src, $dst|$dst, $src}",
4523 [(store (i32 (vector_extract (v4i32 VR128:$src),
4524 (iPTR 0))), addr:$dst)]>;
4526 //===---------------------------------------------------------------------===//
4527 // Move Packed Doubleword Int first element to Doubleword Int
4529 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4530 "mov{d|q}\t{$src, $dst|$dst, $src}",
4531 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4533 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4535 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4536 "mov{d|q}\t{$src, $dst|$dst, $src}",
4537 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4540 //===---------------------------------------------------------------------===//
4541 // Bitcast FR64 <-> GR64
4543 let Predicates = [HasAVX] in
4544 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4545 "vmovq\t{$src, $dst|$dst, $src}",
4546 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4548 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4549 "mov{d|q}\t{$src, $dst|$dst, $src}",
4550 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4551 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4552 "movq\t{$src, $dst|$dst, $src}",
4553 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4555 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4556 "movq\t{$src, $dst|$dst, $src}",
4557 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4558 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4559 "mov{d|q}\t{$src, $dst|$dst, $src}",
4560 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4561 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4562 "movq\t{$src, $dst|$dst, $src}",
4563 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4565 //===---------------------------------------------------------------------===//
4566 // Move Scalar Single to Double Int
4568 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4569 "movd\t{$src, $dst|$dst, $src}",
4570 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4571 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4572 "movd\t{$src, $dst|$dst, $src}",
4573 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4574 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4575 "movd\t{$src, $dst|$dst, $src}",
4576 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4577 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4578 "movd\t{$src, $dst|$dst, $src}",
4579 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4581 //===---------------------------------------------------------------------===//
4582 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4584 let AddedComplexity = 15 in {
4585 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4586 "movd\t{$src, $dst|$dst, $src}",
4587 [(set VR128:$dst, (v4i32 (X86vzmovl
4588 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4590 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4591 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4592 [(set VR128:$dst, (v2i64 (X86vzmovl
4593 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4596 let AddedComplexity = 15 in {
4597 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4598 "movd\t{$src, $dst|$dst, $src}",
4599 [(set VR128:$dst, (v4i32 (X86vzmovl
4600 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4601 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4602 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4603 [(set VR128:$dst, (v2i64 (X86vzmovl
4604 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4607 let AddedComplexity = 20 in {
4608 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4609 "movd\t{$src, $dst|$dst, $src}",
4611 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4612 (loadi32 addr:$src))))))]>,
4614 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4615 "movd\t{$src, $dst|$dst, $src}",
4617 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4618 (loadi32 addr:$src))))))]>;
4621 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4622 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4623 (MOVZDI2PDIrm addr:$src)>;
4624 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4625 (MOVZDI2PDIrm addr:$src)>;
4626 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4627 (MOVZDI2PDIrm addr:$src)>;
4630 let Predicates = [HasAVX] in {
4631 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4632 let AddedComplexity = 20 in {
4633 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4634 (VMOVZDI2PDIrm addr:$src)>;
4635 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4636 (VMOVZDI2PDIrm addr:$src)>;
4637 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4638 (VMOVZDI2PDIrm addr:$src)>;
4640 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4641 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4642 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4643 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4644 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4645 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4646 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4649 // These are the correct encodings of the instructions so that we know how to
4650 // read correct assembly, even though we continue to emit the wrong ones for
4651 // compatibility with Darwin's buggy assembler.
4652 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4653 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4654 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4655 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4656 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4657 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4658 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4659 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4660 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4661 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4662 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4663 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4665 //===---------------------------------------------------------------------===//
4666 // SSE2 - Move Quadword
4667 //===---------------------------------------------------------------------===//
4669 //===---------------------------------------------------------------------===//
4670 // Move Quadword Int to Packed Quadword Int
4672 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4673 "vmovq\t{$src, $dst|$dst, $src}",
4675 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4676 VEX, Requires<[HasAVX]>;
4677 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4678 "movq\t{$src, $dst|$dst, $src}",
4680 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4681 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4683 //===---------------------------------------------------------------------===//
4684 // Move Packed Quadword Int to Quadword Int
4686 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4687 "movq\t{$src, $dst|$dst, $src}",
4688 [(store (i64 (vector_extract (v2i64 VR128:$src),
4689 (iPTR 0))), addr:$dst)]>, VEX;
4690 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4691 "movq\t{$src, $dst|$dst, $src}",
4692 [(store (i64 (vector_extract (v2i64 VR128:$src),
4693 (iPTR 0))), addr:$dst)]>;
4695 //===---------------------------------------------------------------------===//
4696 // Store / copy lower 64-bits of a XMM register.
4698 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4699 "movq\t{$src, $dst|$dst, $src}",
4700 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4701 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4702 "movq\t{$src, $dst|$dst, $src}",
4703 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4705 let AddedComplexity = 20 in
4706 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4707 "vmovq\t{$src, $dst|$dst, $src}",
4709 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4710 (loadi64 addr:$src))))))]>,
4711 XS, VEX, Requires<[HasAVX]>;
4713 let AddedComplexity = 20 in
4714 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4715 "movq\t{$src, $dst|$dst, $src}",
4717 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4718 (loadi64 addr:$src))))))]>,
4719 XS, Requires<[HasSSE2]>;
4721 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4722 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4723 (MOVZQI2PQIrm addr:$src)>;
4724 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4725 (MOVZQI2PQIrm addr:$src)>;
4726 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4729 let Predicates = [HasAVX], AddedComplexity = 20 in {
4730 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4731 (VMOVZQI2PQIrm addr:$src)>;
4732 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4733 (VMOVZQI2PQIrm addr:$src)>;
4734 def : Pat<(v2i64 (X86vzload addr:$src)),
4735 (VMOVZQI2PQIrm addr:$src)>;
4738 let Predicates = [HasAVX] in {
4739 def : Pat<(v4i64 (X86vzload addr:$src)),
4740 (SUBREG_TO_REG (i32 0), (VMOVAPSrm addr:$src), sub_xmm)>;
4743 //===---------------------------------------------------------------------===//
4744 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4745 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4747 let AddedComplexity = 15 in
4748 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4749 "vmovq\t{$src, $dst|$dst, $src}",
4750 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4751 XS, VEX, Requires<[HasAVX]>;
4752 let AddedComplexity = 15 in
4753 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4754 "movq\t{$src, $dst|$dst, $src}",
4755 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4756 XS, Requires<[HasSSE2]>;
4758 let AddedComplexity = 20 in
4759 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4760 "vmovq\t{$src, $dst|$dst, $src}",
4761 [(set VR128:$dst, (v2i64 (X86vzmovl
4762 (loadv2i64 addr:$src))))]>,
4763 XS, VEX, Requires<[HasAVX]>;
4764 let AddedComplexity = 20 in {
4765 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4766 "movq\t{$src, $dst|$dst, $src}",
4767 [(set VR128:$dst, (v2i64 (X86vzmovl
4768 (loadv2i64 addr:$src))))]>,
4769 XS, Requires<[HasSSE2]>;
4772 let AddedComplexity = 20 in {
4773 let Predicates = [HasSSE2] in {
4774 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4775 (MOVZPQILo2PQIrm addr:$src)>;
4776 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4777 (MOVZPQILo2PQIrr VR128:$src)>;
4779 let Predicates = [HasAVX] in {
4780 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4781 (VMOVZPQILo2PQIrm addr:$src)>;
4782 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4783 (VMOVZPQILo2PQIrr VR128:$src)>;
4787 // Instructions to match in the assembler
4788 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4789 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4790 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4791 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4792 // Recognize "movd" with GR64 destination, but encode as a "movq"
4793 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4794 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4796 // Instructions for the disassembler
4797 // xr = XMM register
4800 let Predicates = [HasAVX] in
4801 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4802 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4803 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4804 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4806 //===---------------------------------------------------------------------===//
4807 // SSE3 - Conversion Instructions
4808 //===---------------------------------------------------------------------===//
4810 // Convert Packed Double FP to Packed DW Integers
4811 let Predicates = [HasAVX] in {
4812 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4813 // register, but the same isn't true when using memory operands instead.
4814 // Provide other assembly rr and rm forms to address this explicitly.
4815 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4816 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4817 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4818 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4821 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4822 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4823 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4824 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4827 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4828 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4829 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4830 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4833 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4834 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4835 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4836 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4838 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4839 (VCVTPD2DQYrr VR256:$src)>;
4840 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4841 (VCVTPD2DQYrm addr:$src)>;
4843 // Convert Packed DW Integers to Packed Double FP
4844 let Predicates = [HasAVX] in {
4845 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4846 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4847 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4848 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4849 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4850 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4851 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4852 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4855 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4856 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4857 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4858 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4860 // AVX 256-bit register conversion intrinsics
4861 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4862 (VCVTDQ2PDYrr VR128:$src)>;
4863 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4864 (VCVTDQ2PDYrm addr:$src)>;
4866 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4867 (VCVTPD2DQYrr VR256:$src)>;
4868 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4869 (VCVTPD2DQYrm addr:$src)>;
4871 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4872 (VCVTDQ2PDYrr VR128:$src)>;
4873 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4874 (VCVTDQ2PDYrm addr:$src)>;
4876 //===---------------------------------------------------------------------===//
4877 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4878 //===---------------------------------------------------------------------===//
4879 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4880 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4881 X86MemOperand x86memop> {
4882 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4883 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4884 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4885 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4886 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4887 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4890 let Predicates = [HasAVX] in {
4891 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4892 v4f32, VR128, memopv4f32, f128mem>, VEX;
4893 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4894 v4f32, VR128, memopv4f32, f128mem>, VEX;
4895 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4896 v8f32, VR256, memopv8f32, f256mem>, VEX;
4897 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4898 v8f32, VR256, memopv8f32, f256mem>, VEX;
4900 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4901 memopv4f32, f128mem>;
4902 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4903 memopv4f32, f128mem>;
4905 let Predicates = [HasSSE3] in {
4906 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4907 (MOVSHDUPrr VR128:$src)>;
4908 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4909 (MOVSHDUPrm addr:$src)>;
4910 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4911 (MOVSLDUPrr VR128:$src)>;
4912 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4913 (MOVSLDUPrm addr:$src)>;
4916 let Predicates = [HasAVX] in {
4917 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4918 (VMOVSHDUPrr VR128:$src)>;
4919 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4920 (VMOVSHDUPrm addr:$src)>;
4921 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4922 (VMOVSLDUPrr VR128:$src)>;
4923 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4924 (VMOVSLDUPrm addr:$src)>;
4925 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4926 (VMOVSHDUPYrr VR256:$src)>;
4927 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4928 (VMOVSHDUPYrm addr:$src)>;
4929 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4930 (VMOVSLDUPYrr VR256:$src)>;
4931 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4932 (VMOVSLDUPYrm addr:$src)>;
4935 //===---------------------------------------------------------------------===//
4936 // SSE3 - Replicate Double FP - MOVDDUP
4937 //===---------------------------------------------------------------------===//
4939 multiclass sse3_replicate_dfp<string OpcodeStr> {
4940 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4941 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4942 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4943 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4944 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4946 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4950 // FIXME: Merge with above classe when there're patterns for the ymm version
4951 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4952 let Predicates = [HasAVX] in {
4953 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4954 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4956 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4957 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4962 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4963 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4964 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4966 let Predicates = [HasSSE3] in {
4967 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4969 (MOVDDUPrm addr:$src)>;
4970 let AddedComplexity = 5 in {
4971 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4972 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4973 (MOVDDUPrm addr:$src)>;
4974 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4975 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4976 (MOVDDUPrm addr:$src)>;
4978 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4979 (MOVDDUPrm addr:$src)>;
4980 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4981 (MOVDDUPrm addr:$src)>;
4982 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4983 (MOVDDUPrm addr:$src)>;
4984 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4985 (MOVDDUPrm addr:$src)>;
4986 def : Pat<(X86Movddup (bc_v2f64
4987 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4988 (MOVDDUPrm addr:$src)>;
4991 let Predicates = [HasAVX] in {
4992 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4994 (VMOVDDUPrm addr:$src)>;
4995 let AddedComplexity = 5 in {
4996 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4997 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4998 (VMOVDDUPrm addr:$src)>;
4999 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
5000 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
5001 (VMOVDDUPrm addr:$src)>;
5003 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
5004 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5005 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
5006 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5007 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
5008 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5009 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
5010 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5011 def : Pat<(X86Movddup (bc_v2f64
5012 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
5013 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
5016 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
5017 (VMOVDDUPYrm addr:$src)>;
5018 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
5019 (VMOVDDUPYrm addr:$src)>;
5020 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
5021 (VMOVDDUPYrm addr:$src)>;
5022 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5023 (VMOVDDUPYrm addr:$src)>;
5024 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5025 (VMOVDDUPYrr VR256:$src)>;
5026 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5027 (VMOVDDUPYrr VR256:$src)>;
5030 //===---------------------------------------------------------------------===//
5031 // SSE3 - Move Unaligned Integer
5032 //===---------------------------------------------------------------------===//
5034 let Predicates = [HasAVX] in {
5035 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5036 "vlddqu\t{$src, $dst|$dst, $src}",
5037 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5038 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5039 "vlddqu\t{$src, $dst|$dst, $src}",
5040 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5042 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5043 "lddqu\t{$src, $dst|$dst, $src}",
5044 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5046 //===---------------------------------------------------------------------===//
5047 // SSE3 - Arithmetic
5048 //===---------------------------------------------------------------------===//
5050 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5051 X86MemOperand x86memop, bit Is2Addr = 1> {
5052 def rr : I<0xD0, MRMSrcReg,
5053 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5055 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5056 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5057 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5058 def rm : I<0xD0, MRMSrcMem,
5059 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5061 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5062 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5063 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5066 let Predicates = [HasAVX] in {
5067 let ExeDomain = SSEPackedSingle in {
5068 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5069 f128mem, 0>, TB, XD, VEX_4V;
5070 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5071 f256mem, 0>, TB, XD, VEX_4V;
5073 let ExeDomain = SSEPackedDouble in {
5074 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5075 f128mem, 0>, TB, OpSize, VEX_4V;
5076 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5077 f256mem, 0>, TB, OpSize, VEX_4V;
5080 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5081 let ExeDomain = SSEPackedSingle in
5082 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5084 let ExeDomain = SSEPackedDouble in
5085 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5086 f128mem>, TB, OpSize;
5089 //===---------------------------------------------------------------------===//
5090 // SSE3 Instructions
5091 //===---------------------------------------------------------------------===//
5094 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5095 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5096 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5098 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5099 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5100 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5102 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5104 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5105 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5106 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5108 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5109 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5110 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5112 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5113 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5114 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5116 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5118 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5119 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5120 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5123 let Predicates = [HasAVX] in {
5124 let ExeDomain = SSEPackedSingle in {
5125 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5126 X86fhadd, 0>, VEX_4V;
5127 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5128 X86fhsub, 0>, VEX_4V;
5129 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5130 X86fhadd, 0>, VEX_4V;
5131 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5132 X86fhsub, 0>, VEX_4V;
5134 let ExeDomain = SSEPackedDouble in {
5135 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5136 X86fhadd, 0>, VEX_4V;
5137 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5138 X86fhsub, 0>, VEX_4V;
5139 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5140 X86fhadd, 0>, VEX_4V;
5141 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5142 X86fhsub, 0>, VEX_4V;
5146 let Constraints = "$src1 = $dst" in {
5147 let ExeDomain = SSEPackedSingle in {
5148 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5149 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5151 let ExeDomain = SSEPackedDouble in {
5152 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5153 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5157 //===---------------------------------------------------------------------===//
5158 // SSSE3 - Packed Absolute Instructions
5159 //===---------------------------------------------------------------------===//
5162 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5163 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5164 Intrinsic IntId128> {
5165 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5167 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5168 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5171 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5173 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5176 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5179 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5180 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5181 Intrinsic IntId256> {
5182 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5184 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5185 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5188 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5190 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5193 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5196 let Predicates = [HasAVX] in {
5197 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5198 int_x86_ssse3_pabs_b_128>, VEX;
5199 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5200 int_x86_ssse3_pabs_w_128>, VEX;
5201 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5202 int_x86_ssse3_pabs_d_128>, VEX;
5205 let Predicates = [HasAVX2] in {
5206 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5207 int_x86_avx2_pabs_b>, VEX;
5208 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5209 int_x86_avx2_pabs_w>, VEX;
5210 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5211 int_x86_avx2_pabs_d>, VEX;
5214 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5215 int_x86_ssse3_pabs_b_128>;
5216 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5217 int_x86_ssse3_pabs_w_128>;
5218 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5219 int_x86_ssse3_pabs_d_128>;
5221 //===---------------------------------------------------------------------===//
5222 // SSSE3 - Packed Binary Operator Instructions
5223 //===---------------------------------------------------------------------===//
5225 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5226 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5227 Intrinsic IntId128, bit Is2Addr = 1> {
5228 let isCommutable = 1 in
5229 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5230 (ins VR128:$src1, VR128:$src2),
5232 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5233 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5234 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5236 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5237 (ins VR128:$src1, i128mem:$src2),
5239 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5242 (IntId128 VR128:$src1,
5243 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5246 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5247 Intrinsic IntId256> {
5248 let isCommutable = 1 in
5249 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5250 (ins VR256:$src1, VR256:$src2),
5251 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5252 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5254 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5255 (ins VR256:$src1, i256mem:$src2),
5256 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5258 (IntId256 VR256:$src1,
5259 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5262 let ImmT = NoImm, Predicates = [HasAVX] in {
5263 let isCommutable = 0 in {
5264 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5265 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5266 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5267 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5268 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5269 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5270 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5271 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5272 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5273 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5274 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5275 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5276 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5277 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5278 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5279 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5280 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5281 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5282 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5283 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5284 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5285 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5287 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5288 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5291 let ImmT = NoImm, Predicates = [HasAVX2] in {
5292 let isCommutable = 0 in {
5293 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5294 int_x86_avx2_phadd_w>, VEX_4V;
5295 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5296 int_x86_avx2_phadd_d>, VEX_4V;
5297 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5298 int_x86_avx2_phadd_sw>, VEX_4V;
5299 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5300 int_x86_avx2_phsub_w>, VEX_4V;
5301 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5302 int_x86_avx2_phsub_d>, VEX_4V;
5303 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5304 int_x86_avx2_phsub_sw>, VEX_4V;
5305 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5306 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5307 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5308 int_x86_avx2_pshuf_b>, VEX_4V;
5309 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5310 int_x86_avx2_psign_b>, VEX_4V;
5311 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5312 int_x86_avx2_psign_w>, VEX_4V;
5313 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5314 int_x86_avx2_psign_d>, VEX_4V;
5316 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5317 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5320 // None of these have i8 immediate fields.
5321 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5322 let isCommutable = 0 in {
5323 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5324 int_x86_ssse3_phadd_w_128>;
5325 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5326 int_x86_ssse3_phadd_d_128>;
5327 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5328 int_x86_ssse3_phadd_sw_128>;
5329 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5330 int_x86_ssse3_phsub_w_128>;
5331 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5332 int_x86_ssse3_phsub_d_128>;
5333 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5334 int_x86_ssse3_phsub_sw_128>;
5335 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5336 int_x86_ssse3_pmadd_ub_sw_128>;
5337 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5338 int_x86_ssse3_pshuf_b_128>;
5339 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5340 int_x86_ssse3_psign_b_128>;
5341 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5342 int_x86_ssse3_psign_w_128>;
5343 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5344 int_x86_ssse3_psign_d_128>;
5346 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5347 int_x86_ssse3_pmul_hr_sw_128>;
5350 let Predicates = [HasSSSE3] in {
5351 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5352 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5353 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5354 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5356 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5357 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5358 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5359 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5360 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5361 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5363 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5364 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5365 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5366 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5367 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5368 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5369 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5370 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5373 let Predicates = [HasAVX] in {
5374 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5375 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5376 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5377 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5379 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5380 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5381 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5382 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5383 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5384 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5386 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5387 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5388 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5389 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5390 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5391 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5392 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5393 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5396 let Predicates = [HasAVX2] in {
5397 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5398 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5399 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5400 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5401 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5402 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5404 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5405 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5406 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5407 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5408 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5409 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5410 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5411 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5414 //===---------------------------------------------------------------------===//
5415 // SSSE3 - Packed Align Instruction Patterns
5416 //===---------------------------------------------------------------------===//
5418 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5419 let neverHasSideEffects = 1 in {
5420 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5421 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5423 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5425 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5428 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5429 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5431 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5433 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5438 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5439 let neverHasSideEffects = 1 in {
5440 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5441 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5443 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5446 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5447 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5449 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5454 let Predicates = [HasAVX] in
5455 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5456 let Predicates = [HasAVX2] in
5457 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5458 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5459 defm PALIGN : ssse3_palign<"palignr">;
5461 let Predicates = [HasSSSE3] in {
5462 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5463 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5464 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5465 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5466 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5467 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5468 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5469 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5472 let Predicates = [HasAVX] in {
5473 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5474 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5475 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5476 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5477 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5478 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5479 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5480 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5483 //===---------------------------------------------------------------------===//
5484 // SSSE3 - Thread synchronization
5485 //===---------------------------------------------------------------------===//
5487 let usesCustomInserter = 1 in {
5488 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5489 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>,
5490 Requires<[HasSSE3orAVX]>;
5491 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5492 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>,
5493 Requires<[HasSSE3orAVX]>;
5496 let Uses = [EAX, ECX, EDX] in
5497 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5498 Requires<[HasSSE3orAVX]>;
5499 let Uses = [ECX, EAX] in
5500 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5501 Requires<[HasSSE3orAVX]>;
5503 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5504 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5506 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5507 Requires<[In32BitMode]>;
5508 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5509 Requires<[In64BitMode]>;
5511 //===----------------------------------------------------------------------===//
5512 // SSE4.1 - Packed Move with Sign/Zero Extend
5513 //===----------------------------------------------------------------------===//
5515 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5516 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5517 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5518 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5520 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5521 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5523 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5527 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5529 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5530 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5531 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5533 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5534 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5535 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5538 let Predicates = [HasAVX] in {
5539 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5541 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5543 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5545 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5547 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5549 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5553 let Predicates = [HasAVX2] in {
5554 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5555 int_x86_avx2_pmovsxbw>, VEX;
5556 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5557 int_x86_avx2_pmovsxwd>, VEX;
5558 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5559 int_x86_avx2_pmovsxdq>, VEX;
5560 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5561 int_x86_avx2_pmovzxbw>, VEX;
5562 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5563 int_x86_avx2_pmovzxwd>, VEX;
5564 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5565 int_x86_avx2_pmovzxdq>, VEX;
5568 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5569 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5570 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5571 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5572 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5573 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5575 let Predicates = [HasSSE41] in {
5576 // Common patterns involving scalar load.
5577 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5578 (PMOVSXBWrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5580 (PMOVSXBWrm addr:$src)>;
5582 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5583 (PMOVSXWDrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5585 (PMOVSXWDrm addr:$src)>;
5587 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5588 (PMOVSXDQrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5590 (PMOVSXDQrm addr:$src)>;
5592 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5593 (PMOVZXBWrm addr:$src)>;
5594 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5595 (PMOVZXBWrm addr:$src)>;
5597 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5598 (PMOVZXWDrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5600 (PMOVZXWDrm addr:$src)>;
5602 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5603 (PMOVZXDQrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5605 (PMOVZXDQrm addr:$src)>;
5608 let Predicates = [HasAVX] in {
5609 // Common patterns involving scalar load.
5610 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5611 (VPMOVSXBWrm addr:$src)>;
5612 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5613 (VPMOVSXBWrm addr:$src)>;
5615 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5616 (VPMOVSXWDrm addr:$src)>;
5617 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5618 (VPMOVSXWDrm addr:$src)>;
5620 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5621 (VPMOVSXDQrm addr:$src)>;
5622 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5623 (VPMOVSXDQrm addr:$src)>;
5625 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5626 (VPMOVZXBWrm addr:$src)>;
5627 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5628 (VPMOVZXBWrm addr:$src)>;
5630 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5631 (VPMOVZXWDrm addr:$src)>;
5632 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5633 (VPMOVZXWDrm addr:$src)>;
5635 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5636 (VPMOVZXDQrm addr:$src)>;
5637 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5638 (VPMOVZXDQrm addr:$src)>;
5642 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5643 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5644 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5645 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5647 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5648 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5650 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5654 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5656 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5657 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5658 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5660 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5661 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5663 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5667 let Predicates = [HasAVX] in {
5668 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5670 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5672 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5674 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5678 let Predicates = [HasAVX2] in {
5679 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5680 int_x86_avx2_pmovsxbd>, VEX;
5681 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5682 int_x86_avx2_pmovsxwq>, VEX;
5683 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5684 int_x86_avx2_pmovzxbd>, VEX;
5685 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5686 int_x86_avx2_pmovzxwq>, VEX;
5689 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5690 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5691 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5692 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5694 let Predicates = [HasSSE41] in {
5695 // Common patterns involving scalar load
5696 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5697 (PMOVSXBDrm addr:$src)>;
5698 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5699 (PMOVSXWQrm addr:$src)>;
5701 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5702 (PMOVZXBDrm addr:$src)>;
5703 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5704 (PMOVZXWQrm addr:$src)>;
5707 let Predicates = [HasAVX] in {
5708 // Common patterns involving scalar load
5709 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5710 (VPMOVSXBDrm addr:$src)>;
5711 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5712 (VPMOVSXWQrm addr:$src)>;
5714 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5715 (VPMOVZXBDrm addr:$src)>;
5716 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5717 (VPMOVZXWQrm addr:$src)>;
5720 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5721 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5722 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5723 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5725 // Expecting a i16 load any extended to i32 value.
5726 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5727 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5728 [(set VR128:$dst, (IntId (bitconvert
5729 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5733 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5735 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5736 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5737 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5739 // Expecting a i16 load any extended to i32 value.
5740 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5741 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5742 [(set VR256:$dst, (IntId (bitconvert
5743 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5747 let Predicates = [HasAVX] in {
5748 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5750 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5753 let Predicates = [HasAVX2] in {
5754 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5755 int_x86_avx2_pmovsxbq>, VEX;
5756 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5757 int_x86_avx2_pmovzxbq>, VEX;
5759 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5760 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5762 let Predicates = [HasSSE41] in {
5763 // Common patterns involving scalar load
5764 def : Pat<(int_x86_sse41_pmovsxbq
5765 (bitconvert (v4i32 (X86vzmovl
5766 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5767 (PMOVSXBQrm addr:$src)>;
5769 def : Pat<(int_x86_sse41_pmovzxbq
5770 (bitconvert (v4i32 (X86vzmovl
5771 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5772 (PMOVZXBQrm addr:$src)>;
5775 let Predicates = [HasAVX] in {
5776 // Common patterns involving scalar load
5777 def : Pat<(int_x86_sse41_pmovsxbq
5778 (bitconvert (v4i32 (X86vzmovl
5779 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5780 (VPMOVSXBQrm addr:$src)>;
5782 def : Pat<(int_x86_sse41_pmovzxbq
5783 (bitconvert (v4i32 (X86vzmovl
5784 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5785 (VPMOVZXBQrm addr:$src)>;
5788 //===----------------------------------------------------------------------===//
5789 // SSE4.1 - Extract Instructions
5790 //===----------------------------------------------------------------------===//
5792 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5793 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5794 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5795 (ins VR128:$src1, i32i8imm:$src2),
5796 !strconcat(OpcodeStr,
5797 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5798 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5800 let neverHasSideEffects = 1, mayStore = 1 in
5801 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5802 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5803 !strconcat(OpcodeStr,
5804 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5807 // There's an AssertZext in the way of writing the store pattern
5808 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5811 let Predicates = [HasAVX] in {
5812 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5813 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5814 (ins VR128:$src1, i32i8imm:$src2),
5815 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5818 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5821 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5822 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5823 let neverHasSideEffects = 1, mayStore = 1 in
5824 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5825 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5826 !strconcat(OpcodeStr,
5827 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5830 // There's an AssertZext in the way of writing the store pattern
5831 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5834 let Predicates = [HasAVX] in
5835 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5837 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5840 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5841 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5842 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5843 (ins VR128:$src1, i32i8imm:$src2),
5844 !strconcat(OpcodeStr,
5845 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5847 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5848 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5849 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5850 !strconcat(OpcodeStr,
5851 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5852 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5853 addr:$dst)]>, OpSize;
5856 let Predicates = [HasAVX] in
5857 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5859 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5861 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5862 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5863 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5864 (ins VR128:$src1, i32i8imm:$src2),
5865 !strconcat(OpcodeStr,
5866 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5868 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5869 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5870 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5871 !strconcat(OpcodeStr,
5872 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5873 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5874 addr:$dst)]>, OpSize, REX_W;
5877 let Predicates = [HasAVX] in
5878 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5880 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5882 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5884 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5885 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5886 (ins VR128:$src1, i32i8imm:$src2),
5887 !strconcat(OpcodeStr,
5888 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5890 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5892 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5893 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5894 !strconcat(OpcodeStr,
5895 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5896 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5897 addr:$dst)]>, OpSize;
5900 let ExeDomain = SSEPackedSingle in {
5901 let Predicates = [HasAVX] in {
5902 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5903 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5904 (ins VR128:$src1, i32i8imm:$src2),
5905 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5908 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5911 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5912 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5915 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5916 Requires<[HasSSE41]>;
5917 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5920 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5923 //===----------------------------------------------------------------------===//
5924 // SSE4.1 - Insert Instructions
5925 //===----------------------------------------------------------------------===//
5927 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5928 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5929 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5931 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5933 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5935 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5936 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5937 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5939 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5941 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5943 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5944 imm:$src3))]>, OpSize;
5947 let Predicates = [HasAVX] in
5948 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5949 let Constraints = "$src1 = $dst" in
5950 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5952 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5953 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5954 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5956 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5958 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5960 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5962 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5963 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5965 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5967 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5969 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5970 imm:$src3)))]>, OpSize;
5973 let Predicates = [HasAVX] in
5974 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5975 let Constraints = "$src1 = $dst" in
5976 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5978 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5979 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5980 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5982 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5984 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5986 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5988 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5989 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5991 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5993 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5995 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5996 imm:$src3)))]>, OpSize;
5999 let Predicates = [HasAVX] in
6000 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
6001 let Constraints = "$src1 = $dst" in
6002 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
6004 // insertps has a few different modes, there's the first two here below which
6005 // are optimized inserts that won't zero arbitrary elements in the destination
6006 // vector. The next one matches the intrinsic and could zero arbitrary elements
6007 // in the target vector.
6008 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
6009 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
6010 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
6012 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6014 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6016 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
6018 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
6019 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
6021 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6023 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6025 (X86insrtps VR128:$src1,
6026 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6027 imm:$src3))]>, OpSize;
6030 let ExeDomain = SSEPackedSingle in {
6031 let Constraints = "$src1 = $dst" in
6032 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6033 let Predicates = [HasAVX] in
6034 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6037 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6038 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6040 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6041 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6042 Requires<[HasSSE41]>;
6044 //===----------------------------------------------------------------------===//
6045 // SSE4.1 - Round Instructions
6046 //===----------------------------------------------------------------------===//
6048 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6049 X86MemOperand x86memop, RegisterClass RC,
6050 PatFrag mem_frag32, PatFrag mem_frag64,
6051 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6052 let ExeDomain = SSEPackedSingle in {
6053 // Intrinsic operation, reg.
6054 // Vector intrinsic operation, reg
6055 def PSr : SS4AIi8<opcps, MRMSrcReg,
6056 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6057 !strconcat(OpcodeStr,
6058 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6059 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6062 // Vector intrinsic operation, mem
6063 def PSm : SS4AIi8<opcps, MRMSrcMem,
6064 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6065 !strconcat(OpcodeStr,
6066 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6068 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6070 } // ExeDomain = SSEPackedSingle
6072 let ExeDomain = SSEPackedDouble in {
6073 // Vector intrinsic operation, reg
6074 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6075 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6076 !strconcat(OpcodeStr,
6077 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6078 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6081 // Vector intrinsic operation, mem
6082 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6083 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6084 !strconcat(OpcodeStr,
6085 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6087 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6089 } // ExeDomain = SSEPackedDouble
6092 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6095 Intrinsic F64Int, bit Is2Addr = 1> {
6096 let ExeDomain = GenericDomain in {
6098 def SSr : SS4AIi8<opcss, MRMSrcReg,
6099 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6101 !strconcat(OpcodeStr,
6102 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6103 !strconcat(OpcodeStr,
6104 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6107 // Intrinsic operation, reg.
6108 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6109 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6111 !strconcat(OpcodeStr,
6112 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6113 !strconcat(OpcodeStr,
6114 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6115 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6118 // Intrinsic operation, mem.
6119 def SSm : SS4AIi8<opcss, MRMSrcMem,
6120 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6122 !strconcat(OpcodeStr,
6123 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6124 !strconcat(OpcodeStr,
6125 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6127 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6131 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6132 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6134 !strconcat(OpcodeStr,
6135 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6136 !strconcat(OpcodeStr,
6137 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6140 // Intrinsic operation, reg.
6141 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6142 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6144 !strconcat(OpcodeStr,
6145 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6146 !strconcat(OpcodeStr,
6147 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6148 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6151 // Intrinsic operation, mem.
6152 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6153 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6155 !strconcat(OpcodeStr,
6156 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6157 !strconcat(OpcodeStr,
6158 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6160 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6162 } // ExeDomain = GenericDomain
6165 // FP round - roundss, roundps, roundsd, roundpd
6166 let Predicates = [HasAVX] in {
6168 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6169 memopv4f32, memopv2f64,
6170 int_x86_sse41_round_ps,
6171 int_x86_sse41_round_pd>, VEX;
6172 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6173 memopv8f32, memopv4f64,
6174 int_x86_avx_round_ps_256,
6175 int_x86_avx_round_pd_256>, VEX;
6176 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6177 int_x86_sse41_round_ss,
6178 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6180 def : Pat<(ffloor FR32:$src),
6181 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6182 def : Pat<(f64 (ffloor FR64:$src)),
6183 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6184 def : Pat<(f32 (fnearbyint FR32:$src)),
6185 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6186 def : Pat<(f64 (fnearbyint FR64:$src)),
6187 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6188 def : Pat<(f32 (fceil FR32:$src)),
6189 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6190 def : Pat<(f64 (fceil FR64:$src)),
6191 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6192 def : Pat<(f32 (frint FR32:$src)),
6193 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6194 def : Pat<(f64 (frint FR64:$src)),
6195 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6196 def : Pat<(f32 (ftrunc FR32:$src)),
6197 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6198 def : Pat<(f64 (ftrunc FR64:$src)),
6199 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6202 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6203 memopv4f32, memopv2f64,
6204 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6205 let Constraints = "$src1 = $dst" in
6206 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6207 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6209 def : Pat<(ffloor FR32:$src),
6210 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6211 def : Pat<(f64 (ffloor FR64:$src)),
6212 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6213 def : Pat<(f32 (fnearbyint FR32:$src)),
6214 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6215 def : Pat<(f64 (fnearbyint FR64:$src)),
6216 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6217 def : Pat<(f32 (fceil FR32:$src)),
6218 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6219 def : Pat<(f64 (fceil FR64:$src)),
6220 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6221 def : Pat<(f32 (frint FR32:$src)),
6222 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6223 def : Pat<(f64 (frint FR64:$src)),
6224 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6225 def : Pat<(f32 (ftrunc FR32:$src)),
6226 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6227 def : Pat<(f64 (ftrunc FR64:$src)),
6228 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6230 //===----------------------------------------------------------------------===//
6231 // SSE4.1 - Packed Bit Test
6232 //===----------------------------------------------------------------------===//
6234 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6235 // the intel intrinsic that corresponds to this.
6236 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6237 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6238 "vptest\t{$src2, $src1|$src1, $src2}",
6239 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6241 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6242 "vptest\t{$src2, $src1|$src1, $src2}",
6243 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6246 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6247 "vptest\t{$src2, $src1|$src1, $src2}",
6248 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6250 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6251 "vptest\t{$src2, $src1|$src1, $src2}",
6252 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6256 let Defs = [EFLAGS] in {
6257 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6258 "ptest\t{$src2, $src1|$src1, $src2}",
6259 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6261 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6262 "ptest\t{$src2, $src1|$src1, $src2}",
6263 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6267 // The bit test instructions below are AVX only
6268 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6269 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6270 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6271 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6272 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6273 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6274 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6275 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6279 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6280 let ExeDomain = SSEPackedSingle in {
6281 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6282 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6284 let ExeDomain = SSEPackedDouble in {
6285 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6286 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6290 //===----------------------------------------------------------------------===//
6291 // SSE4.1 - Misc Instructions
6292 //===----------------------------------------------------------------------===//
6294 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6295 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6296 "popcnt{w}\t{$src, $dst|$dst, $src}",
6297 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6299 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6300 "popcnt{w}\t{$src, $dst|$dst, $src}",
6301 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6302 (implicit EFLAGS)]>, OpSize, XS;
6304 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6305 "popcnt{l}\t{$src, $dst|$dst, $src}",
6306 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6308 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6309 "popcnt{l}\t{$src, $dst|$dst, $src}",
6310 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6311 (implicit EFLAGS)]>, XS;
6313 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6314 "popcnt{q}\t{$src, $dst|$dst, $src}",
6315 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6317 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6318 "popcnt{q}\t{$src, $dst|$dst, $src}",
6319 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6320 (implicit EFLAGS)]>, XS;
6325 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6326 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6327 Intrinsic IntId128> {
6328 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6330 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6331 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6332 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6334 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6337 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6340 let Predicates = [HasAVX] in
6341 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6342 int_x86_sse41_phminposuw>, VEX;
6343 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6344 int_x86_sse41_phminposuw>;
6346 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6347 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6348 Intrinsic IntId128, bit Is2Addr = 1> {
6349 let isCommutable = 1 in
6350 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6351 (ins VR128:$src1, VR128:$src2),
6353 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6354 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6355 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6356 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6357 (ins VR128:$src1, i128mem:$src2),
6359 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6362 (IntId128 VR128:$src1,
6363 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6366 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6367 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6368 Intrinsic IntId256> {
6369 let isCommutable = 1 in
6370 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6371 (ins VR256:$src1, VR256:$src2),
6372 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6373 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6374 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6375 (ins VR256:$src1, i256mem:$src2),
6376 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6378 (IntId256 VR256:$src1,
6379 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6382 let Predicates = [HasAVX] in {
6383 let isCommutable = 0 in
6384 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6386 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6388 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6390 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6392 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6394 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6396 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6398 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6400 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6402 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6404 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6407 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6408 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6409 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6410 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6413 let Predicates = [HasAVX2] in {
6414 let isCommutable = 0 in
6415 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6416 int_x86_avx2_packusdw>, VEX_4V;
6417 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6418 int_x86_avx2_pcmpeq_q>, VEX_4V;
6419 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6420 int_x86_avx2_pmins_b>, VEX_4V;
6421 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6422 int_x86_avx2_pmins_d>, VEX_4V;
6423 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6424 int_x86_avx2_pminu_d>, VEX_4V;
6425 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6426 int_x86_avx2_pminu_w>, VEX_4V;
6427 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6428 int_x86_avx2_pmaxs_b>, VEX_4V;
6429 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6430 int_x86_avx2_pmaxs_d>, VEX_4V;
6431 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6432 int_x86_avx2_pmaxu_d>, VEX_4V;
6433 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6434 int_x86_avx2_pmaxu_w>, VEX_4V;
6435 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6436 int_x86_avx2_pmul_dq>, VEX_4V;
6438 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6439 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6440 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6441 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6444 let Constraints = "$src1 = $dst" in {
6445 let isCommutable = 0 in
6446 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6447 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6448 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6449 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6450 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6451 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6452 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6453 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6454 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6455 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6456 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6459 let Predicates = [HasSSE41] in {
6460 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6461 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6462 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6463 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6466 /// SS48I_binop_rm - Simple SSE41 binary operator.
6467 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6468 ValueType OpVT, bit Is2Addr = 1> {
6469 let isCommutable = 1 in
6470 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6471 (ins VR128:$src1, VR128:$src2),
6473 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6475 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6477 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6478 (ins VR128:$src1, i128mem:$src2),
6480 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6481 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6482 [(set VR128:$dst, (OpNode VR128:$src1,
6483 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6487 /// SS48I_binop_rm - Simple SSE41 binary operator.
6488 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6490 let isCommutable = 1 in
6491 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6492 (ins VR256:$src1, VR256:$src2),
6493 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6494 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6496 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6497 (ins VR256:$src1, i256mem:$src2),
6498 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6499 [(set VR256:$dst, (OpNode VR256:$src1,
6500 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6504 let Predicates = [HasAVX] in
6505 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6506 let Predicates = [HasAVX2] in
6507 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6508 let Constraints = "$src1 = $dst" in
6509 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6511 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6512 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6513 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6514 X86MemOperand x86memop, bit Is2Addr = 1> {
6515 let isCommutable = 1 in
6516 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6517 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6519 !strconcat(OpcodeStr,
6520 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6521 !strconcat(OpcodeStr,
6522 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6523 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6525 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6526 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6528 !strconcat(OpcodeStr,
6529 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6530 !strconcat(OpcodeStr,
6531 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6534 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6538 let Predicates = [HasAVX] in {
6539 let isCommutable = 0 in {
6540 let ExeDomain = SSEPackedSingle in {
6541 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6542 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6543 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6544 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6546 let ExeDomain = SSEPackedDouble in {
6547 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6548 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6549 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6550 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6552 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6553 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6554 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6555 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6557 let ExeDomain = SSEPackedSingle in
6558 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6559 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6560 let ExeDomain = SSEPackedDouble in
6561 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6562 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6563 let ExeDomain = SSEPackedSingle in
6564 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6565 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6568 let Predicates = [HasAVX2] in {
6569 let isCommutable = 0 in {
6570 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6571 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6572 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6573 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6577 let Constraints = "$src1 = $dst" in {
6578 let isCommutable = 0 in {
6579 let ExeDomain = SSEPackedSingle in
6580 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6581 VR128, memopv4f32, i128mem>;
6582 let ExeDomain = SSEPackedDouble in
6583 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6584 VR128, memopv2f64, i128mem>;
6585 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6586 VR128, memopv2i64, i128mem>;
6587 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6588 VR128, memopv2i64, i128mem>;
6590 let ExeDomain = SSEPackedSingle in
6591 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6592 VR128, memopv4f32, i128mem>;
6593 let ExeDomain = SSEPackedDouble in
6594 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6595 VR128, memopv2f64, i128mem>;
6598 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6599 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6600 RegisterClass RC, X86MemOperand x86memop,
6601 PatFrag mem_frag, Intrinsic IntId> {
6602 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6603 (ins RC:$src1, RC:$src2, RC:$src3),
6604 !strconcat(OpcodeStr,
6605 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6606 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6607 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6609 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6610 (ins RC:$src1, x86memop:$src2, RC:$src3),
6611 !strconcat(OpcodeStr,
6612 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6614 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6616 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6619 let Predicates = [HasAVX] in {
6620 let ExeDomain = SSEPackedDouble in {
6621 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6622 memopv2f64, int_x86_sse41_blendvpd>;
6623 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6624 memopv4f64, int_x86_avx_blendv_pd_256>;
6625 } // ExeDomain = SSEPackedDouble
6626 let ExeDomain = SSEPackedSingle in {
6627 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6628 memopv4f32, int_x86_sse41_blendvps>;
6629 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6630 memopv8f32, int_x86_avx_blendv_ps_256>;
6631 } // ExeDomain = SSEPackedSingle
6632 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6633 memopv2i64, int_x86_sse41_pblendvb>;
6636 let Predicates = [HasAVX2] in {
6637 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6638 memopv4i64, int_x86_avx2_pblendvb>;
6641 let Predicates = [HasAVX] in {
6642 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6643 (v16i8 VR128:$src2))),
6644 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6645 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6646 (v4i32 VR128:$src2))),
6647 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6648 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6649 (v4f32 VR128:$src2))),
6650 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6651 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6652 (v2i64 VR128:$src2))),
6653 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6654 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6655 (v2f64 VR128:$src2))),
6656 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6657 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6658 (v8i32 VR256:$src2))),
6659 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6660 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6661 (v8f32 VR256:$src2))),
6662 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6663 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6664 (v4i64 VR256:$src2))),
6665 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6666 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6667 (v4f64 VR256:$src2))),
6668 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6671 let Predicates = [HasAVX2] in {
6672 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6673 (v32i8 VR256:$src2))),
6674 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6677 /// SS41I_ternary_int - SSE 4.1 ternary operator
6678 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6679 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6681 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6682 (ins VR128:$src1, VR128:$src2),
6683 !strconcat(OpcodeStr,
6684 "\t{$src2, $dst|$dst, $src2}"),
6685 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6688 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6689 (ins VR128:$src1, i128mem:$src2),
6690 !strconcat(OpcodeStr,
6691 "\t{$src2, $dst|$dst, $src2}"),
6694 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6698 let ExeDomain = SSEPackedDouble in
6699 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6700 int_x86_sse41_blendvpd>;
6701 let ExeDomain = SSEPackedSingle in
6702 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6703 int_x86_sse41_blendvps>;
6704 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6705 int_x86_sse41_pblendvb>;
6707 let Predicates = [HasSSE41] in {
6708 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6709 (v16i8 VR128:$src2))),
6710 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6711 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6712 (v4i32 VR128:$src2))),
6713 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6714 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6715 (v4f32 VR128:$src2))),
6716 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6717 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6718 (v2i64 VR128:$src2))),
6719 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6720 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6721 (v2f64 VR128:$src2))),
6722 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6725 let Predicates = [HasAVX] in
6726 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6727 "vmovntdqa\t{$src, $dst|$dst, $src}",
6728 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6730 let Predicates = [HasAVX2] in
6731 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6732 "vmovntdqa\t{$src, $dst|$dst, $src}",
6733 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6735 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6736 "movntdqa\t{$src, $dst|$dst, $src}",
6737 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6740 //===----------------------------------------------------------------------===//
6741 // SSE4.2 - Compare Instructions
6742 //===----------------------------------------------------------------------===//
6744 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6745 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6746 Intrinsic IntId128, bit Is2Addr = 1> {
6747 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6748 (ins VR128:$src1, VR128:$src2),
6750 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6751 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6752 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6754 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6755 (ins VR128:$src1, i128mem:$src2),
6757 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6758 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6760 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6763 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6764 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6765 Intrinsic IntId256> {
6766 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6767 (ins VR256:$src1, VR256:$src2),
6768 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6769 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6771 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6772 (ins VR256:$src1, i256mem:$src2),
6773 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6775 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6778 let Predicates = [HasAVX] in {
6779 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6782 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6783 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6784 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6785 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6788 let Predicates = [HasAVX2] in {
6789 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6792 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6793 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6794 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6795 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6798 let Constraints = "$src1 = $dst" in
6799 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6801 let Predicates = [HasSSE42] in {
6802 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6803 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6804 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6805 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6808 //===----------------------------------------------------------------------===//
6809 // SSE4.2 - String/text Processing Instructions
6810 //===----------------------------------------------------------------------===//
6812 // Packed Compare Implicit Length Strings, Return Mask
6813 multiclass pseudo_pcmpistrm<string asm> {
6814 def REG : PseudoI<(outs VR128:$dst),
6815 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6816 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6818 def MEM : PseudoI<(outs VR128:$dst),
6819 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6820 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6821 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6824 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6825 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6826 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6829 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6830 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6831 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6832 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6834 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6835 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6836 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6839 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6840 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6841 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6842 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6844 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6845 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6846 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6849 // Packed Compare Explicit Length Strings, Return Mask
6850 multiclass pseudo_pcmpestrm<string asm> {
6851 def REG : PseudoI<(outs VR128:$dst),
6852 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6853 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6854 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6855 def MEM : PseudoI<(outs VR128:$dst),
6856 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6857 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6858 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6861 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6862 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6863 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6866 let Predicates = [HasAVX],
6867 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6868 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6869 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6870 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6872 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6873 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6874 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6877 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6878 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6879 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6880 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6882 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6883 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6884 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6887 // Packed Compare Implicit Length Strings, Return Index
6888 let Defs = [ECX, EFLAGS] in {
6889 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6890 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6891 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6892 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6893 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6894 (implicit EFLAGS)]>, OpSize;
6895 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6896 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6897 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6898 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6899 (implicit EFLAGS)]>, OpSize;
6903 let Predicates = [HasAVX] in {
6904 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6906 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6908 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6910 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6912 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6914 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6918 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6919 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6920 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6921 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6922 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6923 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6925 // Packed Compare Explicit Length Strings, Return Index
6926 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6927 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6928 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6929 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6930 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6931 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6932 (implicit EFLAGS)]>, OpSize;
6933 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6934 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6935 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6937 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6938 (implicit EFLAGS)]>, OpSize;
6942 let Predicates = [HasAVX] in {
6943 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6945 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6947 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6949 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6951 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6953 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6957 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6958 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6959 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6960 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6961 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6962 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6964 //===----------------------------------------------------------------------===//
6965 // SSE4.2 - CRC Instructions
6966 //===----------------------------------------------------------------------===//
6968 // No CRC instructions have AVX equivalents
6970 // crc intrinsic instruction
6971 // This set of instructions are only rm, the only difference is the size
6973 let Constraints = "$src1 = $dst" in {
6974 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6975 (ins GR32:$src1, i8mem:$src2),
6976 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6978 (int_x86_sse42_crc32_32_8 GR32:$src1,
6979 (load addr:$src2)))]>;
6980 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6981 (ins GR32:$src1, GR8:$src2),
6982 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6985 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6986 (ins GR32:$src1, i16mem:$src2),
6987 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6989 (int_x86_sse42_crc32_32_16 GR32:$src1,
6990 (load addr:$src2)))]>,
6992 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6993 (ins GR32:$src1, GR16:$src2),
6994 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6996 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6998 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6999 (ins GR32:$src1, i32mem:$src2),
7000 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7002 (int_x86_sse42_crc32_32_32 GR32:$src1,
7003 (load addr:$src2)))]>;
7004 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
7005 (ins GR32:$src1, GR32:$src2),
7006 "crc32{l} \t{$src2, $src1|$src1, $src2}",
7008 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
7009 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
7010 (ins GR64:$src1, i8mem:$src2),
7011 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7013 (int_x86_sse42_crc32_64_8 GR64:$src1,
7014 (load addr:$src2)))]>,
7016 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
7017 (ins GR64:$src1, GR8:$src2),
7018 "crc32{b} \t{$src2, $src1|$src1, $src2}",
7020 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
7022 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
7023 (ins GR64:$src1, i64mem:$src2),
7024 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7026 (int_x86_sse42_crc32_64_64 GR64:$src1,
7027 (load addr:$src2)))]>,
7029 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7030 (ins GR64:$src1, GR64:$src2),
7031 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7033 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7037 //===----------------------------------------------------------------------===//
7038 // AES-NI Instructions
7039 //===----------------------------------------------------------------------===//
7041 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7042 Intrinsic IntId128, bit Is2Addr = 1> {
7043 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7044 (ins VR128:$src1, VR128:$src2),
7046 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7047 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7048 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7050 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7051 (ins VR128:$src1, i128mem:$src2),
7053 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7054 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7056 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7059 // Perform One Round of an AES Encryption/Decryption Flow
7060 let Predicates = [HasAVX, HasAES] in {
7061 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7062 int_x86_aesni_aesenc, 0>, VEX_4V;
7063 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7064 int_x86_aesni_aesenclast, 0>, VEX_4V;
7065 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7066 int_x86_aesni_aesdec, 0>, VEX_4V;
7067 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7068 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7071 let Constraints = "$src1 = $dst" in {
7072 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7073 int_x86_aesni_aesenc>;
7074 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7075 int_x86_aesni_aesenclast>;
7076 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7077 int_x86_aesni_aesdec>;
7078 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7079 int_x86_aesni_aesdeclast>;
7082 // Perform the AES InvMixColumn Transformation
7083 let Predicates = [HasAVX, HasAES] in {
7084 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7086 "vaesimc\t{$src1, $dst|$dst, $src1}",
7088 (int_x86_aesni_aesimc VR128:$src1))]>,
7090 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7091 (ins i128mem:$src1),
7092 "vaesimc\t{$src1, $dst|$dst, $src1}",
7093 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7096 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7098 "aesimc\t{$src1, $dst|$dst, $src1}",
7100 (int_x86_aesni_aesimc VR128:$src1))]>,
7102 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7103 (ins i128mem:$src1),
7104 "aesimc\t{$src1, $dst|$dst, $src1}",
7105 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7108 // AES Round Key Generation Assist
7109 let Predicates = [HasAVX, HasAES] in {
7110 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7111 (ins VR128:$src1, i8imm:$src2),
7112 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7114 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7116 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7117 (ins i128mem:$src1, i8imm:$src2),
7118 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7120 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7123 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7124 (ins VR128:$src1, i8imm:$src2),
7125 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7127 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7129 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7130 (ins i128mem:$src1, i8imm:$src2),
7131 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7133 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7136 //===----------------------------------------------------------------------===//
7137 // CLMUL Instructions
7138 //===----------------------------------------------------------------------===//
7140 // Carry-less Multiplication instructions
7141 let neverHasSideEffects = 1 in {
7142 let Constraints = "$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, $src3}",
7155 // AVX carry-less Multiplication instructions
7156 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7157 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7158 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7162 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7163 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7164 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7169 multiclass pclmul_alias<string asm, int immop> {
7170 def : InstAlias<!strconcat("pclmul", asm,
7171 "dq {$src, $dst|$dst, $src}"),
7172 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7174 def : InstAlias<!strconcat("pclmul", asm,
7175 "dq {$src, $dst|$dst, $src}"),
7176 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7178 def : InstAlias<!strconcat("vpclmul", asm,
7179 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7180 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7182 def : InstAlias<!strconcat("vpclmul", asm,
7183 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7184 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7186 defm : pclmul_alias<"hqhq", 0x11>;
7187 defm : pclmul_alias<"hqlq", 0x01>;
7188 defm : pclmul_alias<"lqhq", 0x10>;
7189 defm : pclmul_alias<"lqlq", 0x00>;
7191 //===----------------------------------------------------------------------===//
7193 //===----------------------------------------------------------------------===//
7195 //===----------------------------------------------------------------------===//
7196 // VBROADCAST - Load from memory and broadcast to all elements of the
7197 // destination operand
7199 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7200 X86MemOperand x86memop, Intrinsic Int> :
7201 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7202 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7203 [(set RC:$dst, (Int addr:$src))]>, VEX;
7205 // AVX2 adds register forms
7206 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7208 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7209 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7210 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7212 let ExeDomain = SSEPackedSingle in {
7213 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7214 int_x86_avx_vbroadcast_ss>;
7215 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7216 int_x86_avx_vbroadcast_ss_256>;
7218 let ExeDomain = SSEPackedDouble in
7219 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7220 int_x86_avx_vbroadcast_sd_256>;
7221 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7222 int_x86_avx_vbroadcastf128_pd_256>;
7224 let ExeDomain = SSEPackedSingle in {
7225 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7226 int_x86_avx2_vbroadcast_ss_ps>;
7227 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7228 int_x86_avx2_vbroadcast_ss_ps_256>;
7230 let ExeDomain = SSEPackedDouble in
7231 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7232 int_x86_avx2_vbroadcast_sd_pd_256>;
7234 let Predicates = [HasAVX2] in
7235 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7236 int_x86_avx2_vbroadcasti128>;
7238 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7239 (VBROADCASTF128 addr:$src)>;
7242 //===----------------------------------------------------------------------===//
7243 // VINSERTF128 - Insert packed floating-point values
7245 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7246 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7247 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7248 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7251 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7252 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7253 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7257 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7258 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7259 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7260 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7261 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7262 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7264 //===----------------------------------------------------------------------===//
7265 // VEXTRACTF128 - Extract packed floating-point values
7267 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7268 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7269 (ins VR256:$src1, i8imm:$src2),
7270 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7273 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7274 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7275 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7279 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7280 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7281 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7282 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7283 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7284 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7286 //===----------------------------------------------------------------------===//
7287 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7289 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7290 Intrinsic IntLd, Intrinsic IntLd256,
7291 Intrinsic IntSt, Intrinsic IntSt256> {
7292 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7293 (ins VR128:$src1, f128mem:$src2),
7294 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7295 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7297 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7298 (ins VR256:$src1, f256mem:$src2),
7299 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7300 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7302 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7303 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7304 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7305 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7306 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7307 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7308 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7309 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7312 let ExeDomain = SSEPackedSingle in
7313 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7314 int_x86_avx_maskload_ps,
7315 int_x86_avx_maskload_ps_256,
7316 int_x86_avx_maskstore_ps,
7317 int_x86_avx_maskstore_ps_256>;
7318 let ExeDomain = SSEPackedDouble in
7319 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7320 int_x86_avx_maskload_pd,
7321 int_x86_avx_maskload_pd_256,
7322 int_x86_avx_maskstore_pd,
7323 int_x86_avx_maskstore_pd_256>;
7325 //===----------------------------------------------------------------------===//
7326 // VPERMIL - Permute Single and Double Floating-Point Values
7328 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7329 RegisterClass RC, X86MemOperand x86memop_f,
7330 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7331 Intrinsic IntVar, Intrinsic IntImm> {
7332 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7333 (ins RC:$src1, RC:$src2),
7334 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7335 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7336 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7337 (ins RC:$src1, x86memop_i:$src2),
7338 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7339 [(set RC:$dst, (IntVar RC:$src1,
7340 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7342 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7343 (ins RC:$src1, i8imm:$src2),
7344 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7345 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7346 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7347 (ins x86memop_f:$src1, i8imm:$src2),
7348 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7349 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7352 let ExeDomain = SSEPackedSingle in {
7353 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7354 memopv4f32, memopv2i64,
7355 int_x86_avx_vpermilvar_ps,
7356 int_x86_avx_vpermil_ps>;
7357 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7358 memopv8f32, memopv4i64,
7359 int_x86_avx_vpermilvar_ps_256,
7360 int_x86_avx_vpermil_ps_256>;
7362 let ExeDomain = SSEPackedDouble in {
7363 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7364 memopv2f64, memopv2i64,
7365 int_x86_avx_vpermilvar_pd,
7366 int_x86_avx_vpermil_pd>;
7367 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7368 memopv4f64, memopv4i64,
7369 int_x86_avx_vpermilvar_pd_256,
7370 int_x86_avx_vpermil_pd_256>;
7373 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7374 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7375 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7376 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7377 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7378 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7379 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7380 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7381 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7382 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7383 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7384 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7385 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7387 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7388 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7389 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7391 //===----------------------------------------------------------------------===//
7392 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7394 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7395 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7396 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7397 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7400 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7401 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7402 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7406 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7407 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7408 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7409 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7410 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7411 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7413 def : Pat<(int_x86_avx_vperm2f128_ps_256
7414 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7415 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7416 def : Pat<(int_x86_avx_vperm2f128_pd_256
7417 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7418 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7419 def : Pat<(int_x86_avx_vperm2f128_si_256
7420 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7421 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7423 //===----------------------------------------------------------------------===//
7424 // VZERO - Zero YMM registers
7426 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7427 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7428 // Zero All YMM registers
7429 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7430 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7432 // Zero Upper bits of YMM registers
7433 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7434 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7437 //===----------------------------------------------------------------------===//
7438 // Half precision conversion instructions
7439 //===----------------------------------------------------------------------===//
7440 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7441 let Predicates = [HasAVX, HasF16C] in {
7442 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7443 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7444 [(set RC:$dst, (Int VR128:$src))]>,
7446 let neverHasSideEffects = 1, mayLoad = 1 in
7447 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7448 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7452 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7453 let Predicates = [HasAVX, HasF16C] in {
7454 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7455 (ins RC:$src1, i32i8imm:$src2),
7456 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7457 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7459 let neverHasSideEffects = 1, mayLoad = 1 in
7460 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7461 (ins RC:$src1, i32i8imm:$src2),
7462 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7467 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7468 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7469 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7470 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7472 //===----------------------------------------------------------------------===//
7473 // AVX2 Instructions
7474 //===----------------------------------------------------------------------===//
7476 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7477 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7478 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7479 X86MemOperand x86memop> {
7480 let isCommutable = 1 in
7481 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7482 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7483 !strconcat(OpcodeStr,
7484 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7485 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7487 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7488 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7489 !strconcat(OpcodeStr,
7490 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7493 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7497 let isCommutable = 0 in {
7498 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7499 VR128, memopv2i64, i128mem>;
7500 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7501 VR256, memopv4i64, i256mem>;
7504 //===----------------------------------------------------------------------===//
7505 // VPBROADCAST - Load from memory and broadcast to all elements of the
7506 // destination operand
7508 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7509 X86MemOperand x86memop, PatFrag ld_frag,
7510 Intrinsic Int128, Intrinsic Int256> {
7511 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7512 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7513 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7514 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7515 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7517 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7518 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7519 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7520 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7521 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7522 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7524 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7527 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7528 int_x86_avx2_pbroadcastb_128,
7529 int_x86_avx2_pbroadcastb_256>;
7530 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7531 int_x86_avx2_pbroadcastw_128,
7532 int_x86_avx2_pbroadcastw_256>;
7533 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7534 int_x86_avx2_pbroadcastd_128,
7535 int_x86_avx2_pbroadcastd_256>;
7536 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7537 int_x86_avx2_pbroadcastq_128,
7538 int_x86_avx2_pbroadcastq_256>;
7540 let Predicates = [HasAVX2] in {
7541 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7542 (VPBROADCASTBrm addr:$src)>;
7543 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7544 (VPBROADCASTBYrm addr:$src)>;
7545 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7546 (VPBROADCASTWrm addr:$src)>;
7547 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7548 (VPBROADCASTWYrm addr:$src)>;
7549 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7550 (VPBROADCASTDrm addr:$src)>;
7551 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7552 (VPBROADCASTDYrm addr:$src)>;
7553 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7554 (VPBROADCASTQrm addr:$src)>;
7555 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7556 (VPBROADCASTQYrm addr:$src)>;
7559 // AVX1 broadcast patterns
7560 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7561 (VBROADCASTSSYrm addr:$src)>;
7562 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7563 (VBROADCASTSDrm addr:$src)>;
7564 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7565 (VBROADCASTSSYrm addr:$src)>;
7566 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7567 (VBROADCASTSDrm addr:$src)>;
7569 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7570 (VBROADCASTSSrm addr:$src)>;
7571 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7572 (VBROADCASTSSrm addr:$src)>;
7574 //===----------------------------------------------------------------------===//
7575 // VPERM - Permute instructions
7578 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7580 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7581 (ins VR256:$src1, VR256:$src2),
7582 !strconcat(OpcodeStr,
7583 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7584 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7585 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7586 (ins VR256:$src1, i256mem:$src2),
7587 !strconcat(OpcodeStr,
7588 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7589 [(set VR256:$dst, (Int VR256:$src1,
7590 (bitconvert (mem_frag addr:$src2))))]>,
7594 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7595 let ExeDomain = SSEPackedSingle in
7596 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7598 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7600 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7601 (ins VR256:$src1, i8imm:$src2),
7602 !strconcat(OpcodeStr,
7603 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7604 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7605 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7606 (ins i256mem:$src1, i8imm:$src2),
7607 !strconcat(OpcodeStr,
7608 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7609 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7613 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7615 let ExeDomain = SSEPackedDouble in
7616 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7619 //===----------------------------------------------------------------------===//
7620 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7622 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7623 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7624 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7626 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7628 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7629 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7630 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7632 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7636 let Predicates = [HasAVX2] in {
7637 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7638 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7639 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7640 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7641 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7642 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7643 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7644 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7646 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7648 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7649 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7650 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7651 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7652 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7654 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7655 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7657 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7661 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7662 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7663 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7664 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7665 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7666 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7667 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7668 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7669 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7670 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7671 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7672 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7674 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7675 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7676 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7677 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7678 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7679 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7680 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7681 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7682 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7683 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7684 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7685 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7686 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7687 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7688 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7689 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7690 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7691 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7694 //===----------------------------------------------------------------------===//
7695 // VINSERTI128 - Insert packed integer values
7697 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7698 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7699 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7701 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7703 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7704 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7705 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7707 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7708 imm:$src3))]>, VEX_4V;
7710 let Predicates = [HasAVX2] in {
7711 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7713 (VINSERTI128rr VR256:$src1, VR128:$src2,
7714 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7715 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7717 (VINSERTI128rr VR256:$src1, VR128:$src2,
7718 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7719 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7721 (VINSERTI128rr VR256:$src1, VR128:$src2,
7722 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7723 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7725 (VINSERTI128rr VR256:$src1, VR128:$src2,
7726 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7730 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7732 (VINSERTF128rr VR256:$src1, VR128:$src2,
7733 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7734 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7736 (VINSERTF128rr VR256:$src1, VR128:$src2,
7737 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7738 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7740 (VINSERTF128rr VR256:$src1, VR128:$src2,
7741 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7742 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7744 (VINSERTF128rr VR256:$src1, VR128:$src2,
7745 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7746 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7748 (VINSERTF128rr VR256:$src1, VR128:$src2,
7749 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7750 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7752 (VINSERTF128rr VR256:$src1, VR128:$src2,
7753 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7755 //===----------------------------------------------------------------------===//
7756 // VEXTRACTI128 - Extract packed integer values
7758 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7759 (ins VR256:$src1, i8imm:$src2),
7760 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7762 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7764 let neverHasSideEffects = 1, mayStore = 1 in
7765 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7766 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7767 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7769 let Predicates = [HasAVX2] in {
7770 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7771 (v2i64 (VEXTRACTI128rr
7772 (v4i64 VR256:$src1),
7773 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7774 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7775 (v4i32 (VEXTRACTI128rr
7776 (v8i32 VR256:$src1),
7777 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7778 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7779 (v8i16 (VEXTRACTI128rr
7780 (v16i16 VR256:$src1),
7781 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7782 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7783 (v16i8 (VEXTRACTI128rr
7784 (v32i8 VR256:$src1),
7785 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7789 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7790 (v4f32 (VEXTRACTF128rr
7791 (v8f32 VR256:$src1),
7792 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7793 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7794 (v2f64 (VEXTRACTF128rr
7795 (v4f64 VR256:$src1),
7796 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7797 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7798 (v2i64 (VEXTRACTF128rr
7799 (v4i64 VR256:$src1),
7800 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7801 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7802 (v4i32 (VEXTRACTF128rr
7803 (v8i32 VR256:$src1),
7804 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7805 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7806 (v8i16 (VEXTRACTF128rr
7807 (v16i16 VR256:$src1),
7808 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7809 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7810 (v16i8 (VEXTRACTF128rr
7811 (v32i8 VR256:$src1),
7812 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7814 //===----------------------------------------------------------------------===//
7815 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7817 multiclass avx2_pmovmask<string OpcodeStr,
7818 Intrinsic IntLd128, Intrinsic IntLd256,
7819 Intrinsic IntSt128, Intrinsic IntSt256> {
7820 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7821 (ins VR128:$src1, i128mem:$src2),
7822 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7823 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7824 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7825 (ins VR256:$src1, i256mem:$src2),
7826 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7827 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7828 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7829 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7830 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7831 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7832 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7833 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7834 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7835 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7838 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7839 int_x86_avx2_maskload_d,
7840 int_x86_avx2_maskload_d_256,
7841 int_x86_avx2_maskstore_d,
7842 int_x86_avx2_maskstore_d_256>;
7843 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7844 int_x86_avx2_maskload_q,
7845 int_x86_avx2_maskload_q_256,
7846 int_x86_avx2_maskstore_q,
7847 int_x86_avx2_maskstore_q_256>, VEX_W;
7850 //===----------------------------------------------------------------------===//
7851 // Variable Bit Shifts
7853 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7854 ValueType vt128, ValueType vt256> {
7855 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7856 (ins VR128:$src1, VR128:$src2),
7857 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7859 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7861 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7862 (ins VR128:$src1, i128mem:$src2),
7863 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7865 (vt128 (OpNode VR128:$src1,
7866 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7868 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7869 (ins VR256:$src1, VR256:$src2),
7870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7872 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7874 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7875 (ins VR256:$src1, i256mem:$src2),
7876 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7878 (vt256 (OpNode VR256:$src1,
7879 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7883 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7884 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7885 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7886 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7887 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;