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:$src2)>;
1224 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1225 // is during lowering, where it's not possible to recognize the load fold cause
1226 // it has two uses through a bitcast. One use disappears at isel time and the
1227 // fold opportunity reappears.
1228 def : Pat<(v2f64 (X86Unpckl VR128:$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, (undef))), (iPTR 0))), addr:$dst),
1240 (VMOVHPSmr addr:$dst, VR128:$src)>;
1241 def : Pat<(store (f64 (vector_extract
1242 (v2f64 (X86Unpckh VR128:$src, (undef))), (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, (undef))), (iPTR 0))), addr:$dst),
1263 (MOVHPSmr addr:$dst, VR128:$src)>;
1266 let Predicates = [HasSSE2] in {
1267 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1268 // is during lowering, where it's not possible to recognize the load fold cause
1269 // it has two uses through a bitcast. One use disappears at isel time and the
1270 // fold opportunity reappears.
1271 def : Pat<(v2f64 (X86Unpckl VR128:$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, (undef))), (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, $src3}",
2305 memopv2f64, SSEPackedDouble>, TB, OpSize;
2308 let Predicates = [HasSSE1] in {
2309 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2310 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2311 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2312 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2313 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2314 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2315 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2316 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2317 def : Pat<(v4i32 (X86Shufps VR128:$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<(v2f64 (X86Shufps VR128:$src1,
2352 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2353 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2354 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2355 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2356 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2357 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2360 let Predicates = [HasAVX] in {
2361 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2362 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2363 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2364 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2365 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2366 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2367 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2368 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2369 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2370 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2371 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2372 // fall back to this for SSE1)
2373 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2374 (VSHUFPSrri VR128:$src2, VR128:$src1,
2375 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2376 // Special unary SHUFPSrri case.
2377 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2378 (VSHUFPSrri VR128:$src1, VR128:$src1,
2379 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2380 // Special binary v4i32 shuffle cases with SHUFPS.
2381 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2382 (VSHUFPSrri VR128:$src1, VR128:$src2,
2383 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2384 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2385 (bc_v4i32 (memopv2i64 addr:$src2)))),
2386 (VSHUFPSrmi VR128:$src1, addr:$src2,
2387 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2388 // Special unary SHUFPDrri cases.
2389 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2390 (VSHUFPDrri VR128:$src1, VR128:$src1,
2391 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2392 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2393 (VSHUFPDrri VR128:$src1, VR128:$src1,
2394 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2395 // Special binary v2i64 shuffle cases using SHUFPDrri.
2396 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2397 (VSHUFPDrri VR128:$src1, VR128:$src2,
2398 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2400 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2401 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2402 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2403 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2404 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2405 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2406 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2409 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2410 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2411 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2412 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2413 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2415 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2416 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2417 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2418 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2419 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2421 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2422 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2423 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2424 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2425 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2427 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2428 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2429 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2430 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2431 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2434 //===----------------------------------------------------------------------===//
2435 // SSE 1 & 2 - Unpack Instructions
2436 //===----------------------------------------------------------------------===//
2438 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2439 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2440 PatFrag mem_frag, RegisterClass RC,
2441 X86MemOperand x86memop, string asm,
2443 def rr : PI<opc, MRMSrcReg,
2444 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2446 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2447 def rm : PI<opc, MRMSrcMem,
2448 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2450 (vt (OpNode RC:$src1,
2451 (mem_frag addr:$src2))))], d>;
2454 let AddedComplexity = 10 in {
2455 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2456 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2457 SSEPackedSingle>, TB, VEX_4V;
2458 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2459 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2460 SSEPackedDouble>, TB, OpSize, VEX_4V;
2461 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2462 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2463 SSEPackedSingle>, TB, VEX_4V;
2464 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2465 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2466 SSEPackedDouble>, TB, OpSize, VEX_4V;
2468 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2469 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2470 SSEPackedSingle>, TB, VEX_4V;
2471 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2472 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2473 SSEPackedDouble>, TB, OpSize, VEX_4V;
2474 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2475 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2476 SSEPackedSingle>, TB, VEX_4V;
2477 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2478 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2479 SSEPackedDouble>, TB, OpSize, VEX_4V;
2481 let Constraints = "$src1 = $dst" in {
2482 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2483 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2484 SSEPackedSingle>, TB;
2485 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2486 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2487 SSEPackedDouble>, TB, OpSize;
2488 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2489 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2490 SSEPackedSingle>, TB;
2491 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2492 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2493 SSEPackedDouble>, TB, OpSize;
2494 } // Constraints = "$src1 = $dst"
2495 } // AddedComplexity
2497 let Predicates = [HasSSE1] in {
2498 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2499 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2500 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2501 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2502 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2503 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2504 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2505 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2508 let Predicates = [HasSSE2] in {
2509 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2510 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2511 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2512 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2513 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2514 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2515 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2516 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2518 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2519 // problem is during lowering, where it's not possible to recognize the load
2520 // fold cause it has two uses through a bitcast. One use disappears at isel
2521 // time and the fold opportunity reappears.
2522 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2523 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2525 let AddedComplexity = 10 in
2526 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2527 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2530 let Predicates = [HasAVX] in {
2531 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2532 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2533 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2534 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2535 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2536 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2537 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2538 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2540 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2541 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2542 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2543 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2544 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2545 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2546 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2547 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2549 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2550 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2551 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2552 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2553 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2554 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2555 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2556 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2558 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2559 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2560 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2561 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2562 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2563 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2564 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2565 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2567 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2568 // problem is during lowering, where it's not possible to recognize the load
2569 // fold cause it has two uses through a bitcast. One use disappears at isel
2570 // time and the fold opportunity reappears.
2571 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2572 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2573 let AddedComplexity = 10 in
2574 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2575 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2578 //===----------------------------------------------------------------------===//
2579 // SSE 1 & 2 - Extract Floating-Point Sign mask
2580 //===----------------------------------------------------------------------===//
2582 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2583 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2585 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2586 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2587 [(set GR32:$dst, (Int RC:$src))], d>;
2588 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2589 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2592 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2593 SSEPackedSingle>, TB;
2594 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2595 SSEPackedDouble>, TB, OpSize;
2597 def : Pat<(i32 (X86fgetsign FR32:$src)),
2598 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2599 sub_ss))>, Requires<[HasSSE1]>;
2600 def : Pat<(i64 (X86fgetsign FR32:$src)),
2601 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2602 sub_ss))>, Requires<[HasSSE1]>;
2603 def : Pat<(i32 (X86fgetsign FR64:$src)),
2604 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2605 sub_sd))>, Requires<[HasSSE2]>;
2606 def : Pat<(i64 (X86fgetsign FR64:$src)),
2607 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2608 sub_sd))>, Requires<[HasSSE2]>;
2610 let Predicates = [HasAVX] in {
2611 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2612 "movmskps", SSEPackedSingle>, TB, VEX;
2613 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2614 "movmskpd", SSEPackedDouble>, TB,
2616 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2617 "movmskps", SSEPackedSingle>, TB, VEX;
2618 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2619 "movmskpd", SSEPackedDouble>, TB,
2622 def : Pat<(i32 (X86fgetsign FR32:$src)),
2623 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2625 def : Pat<(i64 (X86fgetsign FR32:$src)),
2626 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2628 def : Pat<(i32 (X86fgetsign FR64:$src)),
2629 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2631 def : Pat<(i64 (X86fgetsign FR64:$src)),
2632 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2636 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2637 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2638 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2639 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2641 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2642 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2643 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2644 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2648 //===----------------------------------------------------------------------===//
2649 // SSE 1 & 2 - Logical Instructions
2650 //===----------------------------------------------------------------------===//
2652 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2654 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2656 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2657 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2659 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2660 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2662 let Constraints = "$src1 = $dst" in {
2663 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2664 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2666 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2667 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2671 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2672 let mayLoad = 0 in {
2673 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2674 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2675 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2678 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2679 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2681 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2683 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2685 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2686 // are all promoted to v2i64, and the patterns are covered by the int
2687 // version. This is needed in SSE only, because v2i64 isn't supported on
2688 // SSE1, but only on SSE2.
2689 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2690 !strconcat(OpcodeStr, "ps"), f128mem, [],
2691 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2692 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2694 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2695 !strconcat(OpcodeStr, "pd"), f128mem,
2696 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2697 (bc_v2i64 (v2f64 VR128:$src2))))],
2698 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2699 (memopv2i64 addr:$src2)))], 0>,
2701 let Constraints = "$src1 = $dst" in {
2702 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2703 !strconcat(OpcodeStr, "ps"), f128mem,
2704 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2705 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2706 (memopv2i64 addr:$src2)))]>, TB;
2708 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2709 !strconcat(OpcodeStr, "pd"), f128mem,
2710 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2711 (bc_v2i64 (v2f64 VR128:$src2))))],
2712 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2713 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2717 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2719 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2721 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2722 !strconcat(OpcodeStr, "ps"), f256mem,
2723 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2724 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2725 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2727 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2728 !strconcat(OpcodeStr, "pd"), f256mem,
2729 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2730 (bc_v4i64 (v4f64 VR256:$src2))))],
2731 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2732 (memopv4i64 addr:$src2)))], 0>,
2736 // AVX 256-bit packed logical ops forms
2737 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2738 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2739 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2740 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2742 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2743 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2744 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2745 let isCommutable = 0 in
2746 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2748 //===----------------------------------------------------------------------===//
2749 // SSE 1 & 2 - Arithmetic Instructions
2750 //===----------------------------------------------------------------------===//
2752 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2755 /// In addition, we also have a special variant of the scalar form here to
2756 /// represent the associated intrinsic operation. This form is unlike the
2757 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2758 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2760 /// These three forms can each be reg+reg or reg+mem.
2763 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2765 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2767 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2768 OpNode, FR32, f32mem, Is2Addr>, XS;
2769 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2770 OpNode, FR64, f64mem, Is2Addr>, XD;
2773 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2775 let mayLoad = 0 in {
2776 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2777 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2778 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2779 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2783 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2785 let mayLoad = 0 in {
2786 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2787 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2788 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2789 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2793 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2795 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2796 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2797 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2798 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2801 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2803 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2804 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2805 SSEPackedSingle, Is2Addr>, TB;
2807 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2808 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2809 SSEPackedDouble, Is2Addr>, TB, OpSize;
2812 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2813 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2814 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2815 SSEPackedSingle, 0>, TB;
2817 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2818 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2819 SSEPackedDouble, 0>, TB, OpSize;
2822 // Binary Arithmetic instructions
2823 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2824 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2825 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2826 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2827 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2828 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2829 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2830 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2832 let isCommutable = 0 in {
2833 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2834 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2835 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2836 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2837 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2838 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2839 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2840 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2841 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2842 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2843 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2844 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2845 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2846 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2847 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2848 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2849 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2850 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2851 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2852 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2855 let Constraints = "$src1 = $dst" in {
2856 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2857 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2858 basic_sse12_fp_binop_s_int<0x58, "add">;
2859 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2860 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2861 basic_sse12_fp_binop_s_int<0x59, "mul">;
2863 let isCommutable = 0 in {
2864 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2865 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2866 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2867 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2868 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2869 basic_sse12_fp_binop_s_int<0x5E, "div">;
2870 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2871 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2872 basic_sse12_fp_binop_s_int<0x5F, "max">,
2873 basic_sse12_fp_binop_p_int<0x5F, "max">;
2874 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2875 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2876 basic_sse12_fp_binop_s_int<0x5D, "min">,
2877 basic_sse12_fp_binop_p_int<0x5D, "min">;
2882 /// In addition, we also have a special variant of the scalar form here to
2883 /// represent the associated intrinsic operation. This form is unlike the
2884 /// plain scalar form, in that it takes an entire vector (instead of a
2885 /// scalar) and leaves the top elements undefined.
2887 /// And, we have a special variant form for a full-vector intrinsic form.
2889 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2890 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2891 SDNode OpNode, Intrinsic F32Int> {
2892 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2893 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2894 [(set FR32:$dst, (OpNode FR32:$src))]>;
2895 // For scalar unary operations, fold a load into the operation
2896 // only in OptForSize mode. It eliminates an instruction, but it also
2897 // eliminates a whole-register clobber (the load), so it introduces a
2898 // partial register update condition.
2899 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2900 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2901 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2902 Requires<[HasSSE1, OptForSize]>;
2903 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2904 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2905 [(set VR128:$dst, (F32Int VR128:$src))]>;
2906 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2907 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2908 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2911 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2912 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2913 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2914 !strconcat(OpcodeStr,
2915 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2917 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2918 !strconcat(OpcodeStr,
2919 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2920 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2921 (ins VR128:$src1, ssmem:$src2),
2922 !strconcat(OpcodeStr,
2923 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2926 /// sse1_fp_unop_p - SSE1 unops in packed form.
2927 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2928 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2929 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2930 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2931 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2932 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2933 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2936 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2937 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2938 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2939 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2940 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2941 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2942 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2943 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2946 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2947 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2948 Intrinsic V4F32Int> {
2949 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2950 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2951 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2952 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2953 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2954 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2957 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2958 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2959 Intrinsic V4F32Int> {
2960 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2961 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2962 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2963 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2964 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2965 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2968 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2969 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2970 SDNode OpNode, Intrinsic F64Int> {
2971 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2972 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2973 [(set FR64:$dst, (OpNode FR64:$src))]>;
2974 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2975 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2976 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2977 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2978 Requires<[HasSSE2, OptForSize]>;
2979 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2980 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2981 [(set VR128:$dst, (F64Int VR128:$src))]>;
2982 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2983 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2984 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2987 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2988 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2989 let neverHasSideEffects = 1 in {
2990 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2991 !strconcat(OpcodeStr,
2992 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2994 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2995 !strconcat(OpcodeStr,
2996 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2998 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2999 (ins VR128:$src1, sdmem:$src2),
3000 !strconcat(OpcodeStr,
3001 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3004 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3005 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3007 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3008 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3009 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3010 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3011 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3012 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3015 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3016 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3017 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3018 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3019 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3020 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3021 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3022 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3025 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3026 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3027 Intrinsic V2F64Int> {
3028 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3029 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3030 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3031 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3032 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3033 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3036 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3037 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3038 Intrinsic V2F64Int> {
3039 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3040 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3041 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3042 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3043 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3044 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3047 let Predicates = [HasAVX] in {
3049 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3050 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3052 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3053 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3054 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3055 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3056 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3057 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3058 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3059 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3062 // Reciprocal approximations. Note that these typically require refinement
3063 // in order to obtain suitable precision.
3064 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3065 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3066 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3067 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3068 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3070 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3071 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3072 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3073 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3074 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3077 def : Pat<(f32 (fsqrt FR32:$src)),
3078 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3079 def : Pat<(f32 (fsqrt (load addr:$src))),
3080 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3081 Requires<[HasAVX, OptForSize]>;
3082 def : Pat<(f64 (fsqrt FR64:$src)),
3083 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3084 def : Pat<(f64 (fsqrt (load addr:$src))),
3085 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3086 Requires<[HasAVX, OptForSize]>;
3088 def : Pat<(f32 (X86frsqrt FR32:$src)),
3089 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3090 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3091 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3092 Requires<[HasAVX, OptForSize]>;
3094 def : Pat<(f32 (X86frcp FR32:$src)),
3095 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3096 def : Pat<(f32 (X86frcp (load addr:$src))),
3097 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3098 Requires<[HasAVX, OptForSize]>;
3100 let Predicates = [HasAVX] in {
3101 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3102 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3103 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3104 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3106 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3107 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3109 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3110 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3111 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3112 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3114 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3115 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3117 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3118 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3119 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3120 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3122 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3123 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3125 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3126 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3127 (VRCPSSr (f32 (IMPLICIT_DEF)),
3128 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3130 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3131 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3135 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3136 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3137 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3138 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3139 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3140 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3142 // Reciprocal approximations. Note that these typically require refinement
3143 // in order to obtain suitable precision.
3144 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3145 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3146 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3147 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3148 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3149 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3151 // There is no f64 version of the reciprocal approximation instructions.
3153 //===----------------------------------------------------------------------===//
3154 // SSE 1 & 2 - Non-temporal stores
3155 //===----------------------------------------------------------------------===//
3157 let AddedComplexity = 400 in { // Prefer non-temporal versions
3158 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3159 (ins f128mem:$dst, VR128:$src),
3160 "movntps\t{$src, $dst|$dst, $src}",
3161 [(alignednontemporalstore (v4f32 VR128:$src),
3163 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3164 (ins f128mem:$dst, VR128:$src),
3165 "movntpd\t{$src, $dst|$dst, $src}",
3166 [(alignednontemporalstore (v2f64 VR128:$src),
3168 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3169 (ins f128mem:$dst, VR128:$src),
3170 "movntdq\t{$src, $dst|$dst, $src}",
3171 [(alignednontemporalstore (v2f64 VR128:$src),
3174 let ExeDomain = SSEPackedInt in
3175 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3176 (ins f128mem:$dst, VR128:$src),
3177 "movntdq\t{$src, $dst|$dst, $src}",
3178 [(alignednontemporalstore (v4f32 VR128:$src),
3181 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3182 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3184 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3185 (ins f256mem:$dst, VR256:$src),
3186 "movntps\t{$src, $dst|$dst, $src}",
3187 [(alignednontemporalstore (v8f32 VR256:$src),
3189 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3190 (ins f256mem:$dst, VR256:$src),
3191 "movntpd\t{$src, $dst|$dst, $src}",
3192 [(alignednontemporalstore (v4f64 VR256:$src),
3194 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3195 (ins f256mem:$dst, VR256:$src),
3196 "movntdq\t{$src, $dst|$dst, $src}",
3197 [(alignednontemporalstore (v4f64 VR256:$src),
3199 let ExeDomain = SSEPackedInt in
3200 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3201 (ins f256mem:$dst, VR256:$src),
3202 "movntdq\t{$src, $dst|$dst, $src}",
3203 [(alignednontemporalstore (v8f32 VR256:$src),
3207 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3208 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3209 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3210 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3211 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3212 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3214 let AddedComplexity = 400 in { // Prefer non-temporal versions
3215 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3216 "movntps\t{$src, $dst|$dst, $src}",
3217 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3218 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3219 "movntpd\t{$src, $dst|$dst, $src}",
3220 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3222 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3223 "movntdq\t{$src, $dst|$dst, $src}",
3224 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3226 let ExeDomain = SSEPackedInt in
3227 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3228 "movntdq\t{$src, $dst|$dst, $src}",
3229 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3231 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3232 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3234 // There is no AVX form for instructions below this point
3235 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3236 "movnti{l}\t{$src, $dst|$dst, $src}",
3237 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3238 TB, Requires<[HasSSE2]>;
3239 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3240 "movnti{q}\t{$src, $dst|$dst, $src}",
3241 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3242 TB, Requires<[HasSSE2]>;
3245 //===----------------------------------------------------------------------===//
3246 // SSE 1 & 2 - Prefetch and memory fence
3247 //===----------------------------------------------------------------------===//
3249 // Prefetch intrinsic.
3250 def PREFETCHT0 : VoPSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3251 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3252 def PREFETCHT1 : VoPSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3253 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3254 def PREFETCHT2 : VoPSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3255 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3256 def PREFETCHNTA : VoPSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3257 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3260 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3261 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3262 TB, Requires<[HasSSE2]>;
3264 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3265 // was introduced with SSE2, it's backward compatible.
3266 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3268 // Load, store, and memory fence
3269 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3270 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3271 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3272 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3273 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3274 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3276 def : Pat<(X86SFence), (SFENCE)>;
3277 def : Pat<(X86LFence), (LFENCE)>;
3278 def : Pat<(X86MFence), (MFENCE)>;
3280 //===----------------------------------------------------------------------===//
3281 // SSE 1 & 2 - Load/Store XCSR register
3282 //===----------------------------------------------------------------------===//
3284 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3285 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3286 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3287 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3289 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3290 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3291 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3292 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3294 //===---------------------------------------------------------------------===//
3295 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3296 //===---------------------------------------------------------------------===//
3298 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3300 let neverHasSideEffects = 1 in {
3301 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3302 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3303 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3304 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3306 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3307 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3308 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3309 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3312 let isCodeGenOnly = 1 in {
3313 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3314 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3315 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3316 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3317 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3318 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3319 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3320 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3323 let canFoldAsLoad = 1, mayLoad = 1 in {
3324 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3325 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3326 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3327 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3328 let Predicates = [HasAVX] in {
3329 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3330 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3331 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3332 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3336 let mayStore = 1 in {
3337 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3338 (ins i128mem:$dst, VR128:$src),
3339 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3340 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3341 (ins i256mem:$dst, VR256:$src),
3342 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3343 let Predicates = [HasAVX] in {
3344 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3345 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3346 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3347 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3351 let neverHasSideEffects = 1 in
3352 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3353 "movdqa\t{$src, $dst|$dst, $src}", []>;
3355 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3356 "movdqu\t{$src, $dst|$dst, $src}",
3357 []>, XS, Requires<[HasSSE2]>;
3360 let isCodeGenOnly = 1 in {
3361 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3362 "movdqa\t{$src, $dst|$dst, $src}", []>;
3364 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3365 "movdqu\t{$src, $dst|$dst, $src}",
3366 []>, XS, Requires<[HasSSE2]>;
3369 let canFoldAsLoad = 1, mayLoad = 1 in {
3370 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3371 "movdqa\t{$src, $dst|$dst, $src}",
3372 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3373 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3374 "movdqu\t{$src, $dst|$dst, $src}",
3375 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3376 XS, Requires<[HasSSE2]>;
3379 let mayStore = 1 in {
3380 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3381 "movdqa\t{$src, $dst|$dst, $src}",
3382 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3383 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3384 "movdqu\t{$src, $dst|$dst, $src}",
3385 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3386 XS, Requires<[HasSSE2]>;
3389 // Intrinsic forms of MOVDQU load and store
3390 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3391 "vmovdqu\t{$src, $dst|$dst, $src}",
3392 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3393 XS, VEX, Requires<[HasAVX]>;
3395 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3396 "movdqu\t{$src, $dst|$dst, $src}",
3397 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3398 XS, Requires<[HasSSE2]>;
3400 } // ExeDomain = SSEPackedInt
3402 let Predicates = [HasAVX] in {
3403 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3404 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3405 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3408 //===---------------------------------------------------------------------===//
3409 // SSE2 - Packed Integer Arithmetic Instructions
3410 //===---------------------------------------------------------------------===//
3412 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3414 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3415 RegisterClass RC, PatFrag memop_frag,
3416 X86MemOperand x86memop, bit IsCommutable = 0,
3418 let isCommutable = IsCommutable in
3419 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3420 (ins RC:$src1, RC:$src2),
3422 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3423 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3424 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3425 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3426 (ins RC:$src1, x86memop:$src2),
3428 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3429 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3430 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3433 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3434 string OpcodeStr, Intrinsic IntId,
3435 Intrinsic IntId2, RegisterClass RC,
3437 // src2 is always 128-bit
3438 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3439 (ins RC:$src1, VR128:$src2),
3441 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3442 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3443 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3444 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3445 (ins RC:$src1, i128mem:$src2),
3447 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3449 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3450 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3451 (ins RC:$src1, i32i8imm:$src2),
3453 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3454 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3455 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3458 /// PDI_binop_rm - Simple SSE2 binary operator.
3459 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3460 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
3461 X86MemOperand x86memop, bit IsCommutable = 0,
3463 let isCommutable = IsCommutable in
3464 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3465 (ins RC:$src1, RC:$src2),
3467 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3469 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
3470 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3471 (ins RC:$src1, x86memop:$src2),
3473 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3475 [(set RC:$dst, (OpVT (OpNode RC:$src1,
3476 (bitconvert (memop_frag addr:$src2)))))]>;
3478 } // ExeDomain = SSEPackedInt
3480 // 128-bit Integer Arithmetic
3482 let Predicates = [HasAVX] in {
3483 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3484 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3485 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3486 i128mem, 1, 0>, VEX_4V;
3487 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3488 i128mem, 1, 0>, VEX_4V;
3489 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3490 i128mem, 1, 0>, VEX_4V;
3491 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3492 i128mem, 1, 0>, VEX_4V;
3493 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3494 i128mem, 0, 0>, VEX_4V;
3495 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3496 i128mem, 0, 0>, VEX_4V;
3497 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3498 i128mem, 0, 0>, VEX_4V;
3499 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3500 i128mem, 0, 0>, VEX_4V;
3503 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3504 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3505 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3506 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3507 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3508 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3509 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3510 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3511 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3512 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3513 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3514 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3515 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3516 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3517 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3518 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3519 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3520 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3521 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3522 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3523 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3524 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3525 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3526 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3527 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3528 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3529 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3530 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3531 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3532 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3533 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3534 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3535 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3536 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3537 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3538 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3539 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3540 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3543 let Predicates = [HasAVX2] in {
3544 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3545 i256mem, 1, 0>, VEX_4V;
3546 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3547 i256mem, 1, 0>, VEX_4V;
3548 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3549 i256mem, 1, 0>, VEX_4V;
3550 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3551 i256mem, 1, 0>, VEX_4V;
3552 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3553 i256mem, 1, 0>, VEX_4V;
3554 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3555 i256mem, 0, 0>, VEX_4V;
3556 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3557 i256mem, 0, 0>, VEX_4V;
3558 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3559 i256mem, 0, 0>, VEX_4V;
3560 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3561 i256mem, 0, 0>, VEX_4V;
3564 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3565 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3566 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3567 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3568 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3569 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3570 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3571 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3572 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3573 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3574 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3575 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3576 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3577 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3578 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3579 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3580 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3581 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3582 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3583 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3584 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3585 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3586 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3587 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3588 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3589 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3590 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3591 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3592 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3593 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3594 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3595 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3596 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3597 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3598 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3599 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3600 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3601 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3604 let Constraints = "$src1 = $dst" in {
3605 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3607 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3609 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3611 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3613 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3615 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3617 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3619 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3621 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3625 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3626 VR128, memopv2i64, i128mem>;
3627 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3628 VR128, memopv2i64, i128mem>;
3629 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3630 VR128, memopv2i64, i128mem>;
3631 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3632 VR128, memopv2i64, i128mem>;
3633 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3634 VR128, memopv2i64, i128mem, 1>;
3635 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3636 VR128, memopv2i64, i128mem, 1>;
3637 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3638 VR128, memopv2i64, i128mem, 1>;
3639 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3640 VR128, memopv2i64, i128mem, 1>;
3641 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3642 VR128, memopv2i64, i128mem, 1>;
3643 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3644 VR128, memopv2i64, i128mem, 1>;
3645 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3646 VR128, memopv2i64, i128mem, 1>;
3647 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3648 VR128, memopv2i64, i128mem, 1>;
3649 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3650 VR128, memopv2i64, i128mem, 1>;
3651 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3652 VR128, memopv2i64, i128mem, 1>;
3653 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3654 VR128, memopv2i64, i128mem, 1>;
3655 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3656 VR128, memopv2i64, i128mem, 1>;
3657 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3658 VR128, memopv2i64, i128mem, 1>;
3659 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3660 VR128, memopv2i64, i128mem, 1>;
3661 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3662 VR128, memopv2i64, i128mem, 1>;
3664 } // Constraints = "$src1 = $dst"
3666 //===---------------------------------------------------------------------===//
3667 // SSE2 - Packed Integer Logical Instructions
3668 //===---------------------------------------------------------------------===//
3670 let Predicates = [HasAVX] in {
3671 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3672 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3674 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3675 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3677 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3678 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3681 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3682 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3684 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3685 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3687 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3688 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3691 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3692 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3694 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3695 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3698 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
3699 i128mem, 1, 0>, VEX_4V;
3700 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
3701 i128mem, 1, 0>, VEX_4V;
3702 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
3703 i128mem, 1, 0>, VEX_4V;
3704 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
3705 i128mem, 0, 0>, VEX_4V;
3707 let ExeDomain = SSEPackedInt in {
3708 let neverHasSideEffects = 1 in {
3709 // 128-bit logical shifts.
3710 def VPSLLDQri : PDIi8<0x73, MRM7r,
3711 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3712 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3714 def VPSRLDQri : PDIi8<0x73, MRM3r,
3715 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3716 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3718 // PSRADQri doesn't exist in SSE[1-3].
3723 let Predicates = [HasAVX2] in {
3724 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3725 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3727 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3728 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3730 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3731 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3734 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3735 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3737 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3738 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3740 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3741 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3744 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3745 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3747 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3748 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3751 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
3752 i256mem, 1, 0>, VEX_4V;
3753 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
3754 i256mem, 1, 0>, VEX_4V;
3755 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
3756 i256mem, 1, 0>, VEX_4V;
3757 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
3758 i256mem, 0, 0>, VEX_4V;
3760 let ExeDomain = SSEPackedInt in {
3761 let neverHasSideEffects = 1 in {
3762 // 128-bit logical shifts.
3763 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3764 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3765 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3767 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3768 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3769 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3771 // PSRADQYri doesn't exist in SSE[1-3].
3776 let Constraints = "$src1 = $dst" in {
3777 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3778 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3780 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3781 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3783 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3784 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3787 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3788 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3790 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3791 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3793 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3794 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3797 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3798 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3800 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3801 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3804 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
3806 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
3808 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
3810 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
3813 let ExeDomain = SSEPackedInt in {
3814 let neverHasSideEffects = 1 in {
3815 // 128-bit logical shifts.
3816 def PSLLDQri : PDIi8<0x73, MRM7r,
3817 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3818 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3819 def PSRLDQri : PDIi8<0x73, MRM3r,
3820 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3821 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3822 // PSRADQri doesn't exist in SSE[1-3].
3825 } // Constraints = "$src1 = $dst"
3827 let Predicates = [HasAVX] in {
3828 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3829 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3830 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3831 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3832 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3833 (VPSLLDQri VR128:$src1, imm:$src2)>;
3834 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3835 (VPSRLDQri VR128:$src1, imm:$src2)>;
3836 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3837 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3839 // Shift up / down and insert zero's.
3840 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3841 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3842 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3843 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3846 let Predicates = [HasAVX2] in {
3847 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3848 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3849 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3850 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3851 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3852 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3853 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3854 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3857 let Predicates = [HasSSE2] in {
3858 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3859 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3860 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3861 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3862 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3863 (PSLLDQri VR128:$src1, imm:$src2)>;
3864 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3865 (PSRLDQri VR128:$src1, imm:$src2)>;
3866 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3867 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3869 // Shift up / down and insert zero's.
3870 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3871 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3872 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3873 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3876 //===---------------------------------------------------------------------===//
3877 // SSE2 - Packed Integer Comparison Instructions
3878 //===---------------------------------------------------------------------===//
3880 let Predicates = [HasAVX] in {
3881 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3882 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3883 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3884 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3885 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3886 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3887 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3888 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3889 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3890 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3891 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3892 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3894 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3895 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3896 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3897 (bc_v16i8 (memopv2i64 addr:$src2)))),
3898 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3899 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3900 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3901 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3902 (bc_v8i16 (memopv2i64 addr:$src2)))),
3903 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3904 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3905 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3906 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3907 (bc_v4i32 (memopv2i64 addr:$src2)))),
3908 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3910 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3911 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3912 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3913 (bc_v16i8 (memopv2i64 addr:$src2)))),
3914 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3915 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3916 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3917 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3918 (bc_v8i16 (memopv2i64 addr:$src2)))),
3919 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3920 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3921 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3922 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3923 (bc_v4i32 (memopv2i64 addr:$src2)))),
3924 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3927 let Predicates = [HasAVX2] in {
3928 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3929 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3930 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3931 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3932 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3933 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3934 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3935 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3936 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3937 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3938 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3939 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3941 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3942 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3943 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3944 (bc_v32i8 (memopv4i64 addr:$src2)))),
3945 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3946 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3947 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3948 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3949 (bc_v16i16 (memopv4i64 addr:$src2)))),
3950 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3951 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3952 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3953 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3954 (bc_v8i32 (memopv4i64 addr:$src2)))),
3955 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3957 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3958 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3959 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3960 (bc_v32i8 (memopv4i64 addr:$src2)))),
3961 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3962 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3963 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3964 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3965 (bc_v16i16 (memopv4i64 addr:$src2)))),
3966 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3967 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3968 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3969 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3970 (bc_v8i32 (memopv4i64 addr:$src2)))),
3971 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3974 let Constraints = "$src1 = $dst" in {
3975 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3976 VR128, memopv2i64, i128mem, 1>;
3977 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3978 VR128, memopv2i64, i128mem, 1>;
3979 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3980 VR128, memopv2i64, i128mem, 1>;
3981 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3982 VR128, memopv2i64, i128mem>;
3983 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3984 VR128, memopv2i64, i128mem>;
3985 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3986 VR128, memopv2i64, i128mem>;
3987 } // Constraints = "$src1 = $dst"
3989 let Predicates = [HasSSE2] in {
3990 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3991 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3992 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3993 (bc_v16i8 (memopv2i64 addr:$src2)))),
3994 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3995 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3996 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3997 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3998 (bc_v8i16 (memopv2i64 addr:$src2)))),
3999 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4000 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4001 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4002 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4003 (bc_v4i32 (memopv2i64 addr:$src2)))),
4004 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4006 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4007 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4008 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4009 (bc_v16i8 (memopv2i64 addr:$src2)))),
4010 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4011 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4012 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4013 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4014 (bc_v8i16 (memopv2i64 addr:$src2)))),
4015 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4016 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4017 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4018 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4019 (bc_v4i32 (memopv2i64 addr:$src2)))),
4020 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4023 //===---------------------------------------------------------------------===//
4024 // SSE2 - Packed Integer Pack Instructions
4025 //===---------------------------------------------------------------------===//
4027 let Predicates = [HasAVX] in {
4028 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4029 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4030 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4031 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4032 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4033 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4036 let Predicates = [HasAVX2] in {
4037 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4038 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4039 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4040 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4041 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4042 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4045 let Constraints = "$src1 = $dst" in {
4046 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4047 VR128, memopv2i64, i128mem>;
4048 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4049 VR128, memopv2i64, i128mem>;
4050 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4051 VR128, memopv2i64, i128mem>;
4052 } // Constraints = "$src1 = $dst"
4054 //===---------------------------------------------------------------------===//
4055 // SSE2 - Packed Integer Shuffle Instructions
4056 //===---------------------------------------------------------------------===//
4058 let ExeDomain = SSEPackedInt in {
4059 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4061 def ri : Ii8<0x70, MRMSrcReg,
4062 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4063 !strconcat(OpcodeStr,
4064 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4065 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4067 def mi : Ii8<0x70, MRMSrcMem,
4068 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4069 !strconcat(OpcodeStr,
4070 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4071 [(set VR128:$dst, (vt (pshuf_frag:$src2
4072 (bc_frag (memopv2i64 addr:$src1)),
4076 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4078 def Yri : Ii8<0x70, MRMSrcReg,
4079 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4080 !strconcat(OpcodeStr,
4081 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4082 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4084 def Ymi : Ii8<0x70, MRMSrcMem,
4085 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4086 !strconcat(OpcodeStr,
4087 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4088 [(set VR256:$dst, (vt (pshuf_frag:$src2
4089 (bc_frag (memopv4i64 addr:$src1)),
4092 } // ExeDomain = SSEPackedInt
4094 let Predicates = [HasAVX] in {
4095 let AddedComplexity = 5 in
4096 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4099 // SSE2 with ImmT == Imm8 and XS prefix.
4100 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4103 // SSE2 with ImmT == Imm8 and XD prefix.
4104 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4107 let AddedComplexity = 5 in
4108 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4109 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4110 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4111 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4112 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4114 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4116 (VPSHUFDmi addr:$src1, imm:$imm)>;
4117 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4119 (VPSHUFDmi addr:$src1, imm:$imm)>;
4120 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4121 (VPSHUFDri VR128:$src1, imm:$imm)>;
4122 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4123 (VPSHUFDri VR128:$src1, imm:$imm)>;
4124 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4125 (VPSHUFHWri VR128:$src, imm:$imm)>;
4126 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4128 (VPSHUFHWmi addr:$src, imm:$imm)>;
4129 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4130 (VPSHUFLWri VR128:$src, imm:$imm)>;
4131 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4133 (VPSHUFLWmi addr:$src, imm:$imm)>;
4136 let Predicates = [HasAVX2] in {
4137 let AddedComplexity = 5 in
4138 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4141 // SSE2 with ImmT == Imm8 and XS prefix.
4142 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4145 // SSE2 with ImmT == Imm8 and XD prefix.
4146 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4150 let Predicates = [HasSSE2] in {
4151 let AddedComplexity = 5 in
4152 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4154 // SSE2 with ImmT == Imm8 and XS prefix.
4155 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4157 // SSE2 with ImmT == Imm8 and XD prefix.
4158 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4160 let AddedComplexity = 5 in
4161 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4162 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4163 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4164 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4165 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4167 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4169 (PSHUFDmi addr:$src1, imm:$imm)>;
4170 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4172 (PSHUFDmi addr:$src1, imm:$imm)>;
4173 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4174 (PSHUFDri VR128:$src1, imm:$imm)>;
4175 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4176 (PSHUFDri VR128:$src1, imm:$imm)>;
4177 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4178 (PSHUFHWri VR128:$src, imm:$imm)>;
4179 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4181 (PSHUFHWmi addr:$src, imm:$imm)>;
4182 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4183 (PSHUFLWri VR128:$src, imm:$imm)>;
4184 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4186 (PSHUFLWmi addr:$src, imm:$imm)>;
4189 //===---------------------------------------------------------------------===//
4190 // SSE2 - Packed Integer Unpack Instructions
4191 //===---------------------------------------------------------------------===//
4193 let ExeDomain = SSEPackedInt in {
4194 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4195 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4196 def rr : PDI<opc, MRMSrcReg,
4197 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4199 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4200 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4201 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4202 def rm : PDI<opc, MRMSrcMem,
4203 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4205 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4206 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4207 [(set VR128:$dst, (OpNode VR128:$src1,
4208 (bc_frag (memopv2i64
4212 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4213 SDNode OpNode, PatFrag bc_frag> {
4214 def Yrr : PDI<opc, MRMSrcReg,
4215 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4216 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4217 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4218 def Yrm : PDI<opc, MRMSrcMem,
4219 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4220 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4221 [(set VR256:$dst, (OpNode VR256:$src1,
4222 (bc_frag (memopv4i64 addr:$src2))))]>;
4225 let Predicates = [HasAVX] in {
4226 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4227 bc_v16i8, 0>, VEX_4V;
4228 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4229 bc_v8i16, 0>, VEX_4V;
4230 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4231 bc_v4i32, 0>, VEX_4V;
4232 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4233 bc_v2i64, 0>, VEX_4V;
4235 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4236 bc_v16i8, 0>, VEX_4V;
4237 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4238 bc_v8i16, 0>, VEX_4V;
4239 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4240 bc_v4i32, 0>, VEX_4V;
4241 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4242 bc_v2i64, 0>, VEX_4V;
4245 let Predicates = [HasAVX2] in {
4246 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4248 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4250 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4252 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4255 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4257 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4259 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4261 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4265 let Constraints = "$src1 = $dst" in {
4266 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4268 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4270 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4272 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4275 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4277 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4279 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4281 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4284 } // ExeDomain = SSEPackedInt
4286 // Patterns for using AVX1 instructions with integer vectors
4287 // Here to give AVX2 priority
4288 let Predicates = [HasAVX] in {
4289 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4290 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4291 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4292 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4293 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4294 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4295 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4296 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4298 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4299 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4300 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4301 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4302 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4303 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4304 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4305 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4308 // Splat v2f64 / v2i64
4309 let AddedComplexity = 10 in {
4310 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4311 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4312 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4313 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4316 //===---------------------------------------------------------------------===//
4317 // SSE2 - Packed Integer Extract and Insert
4318 //===---------------------------------------------------------------------===//
4320 let ExeDomain = SSEPackedInt in {
4321 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4322 def rri : Ii8<0xC4, MRMSrcReg,
4323 (outs VR128:$dst), (ins VR128:$src1,
4324 GR32:$src2, i32i8imm:$src3),
4326 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4327 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4329 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4330 def rmi : Ii8<0xC4, MRMSrcMem,
4331 (outs VR128:$dst), (ins VR128:$src1,
4332 i16mem:$src2, i32i8imm:$src3),
4334 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4335 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4337 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4342 let Predicates = [HasAVX] in
4343 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4344 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4345 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4346 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4347 imm:$src2))]>, TB, OpSize, VEX;
4348 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4349 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4350 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4351 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4355 let Predicates = [HasAVX] in {
4356 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4357 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4358 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4359 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4360 []>, TB, OpSize, VEX_4V;
4363 let Constraints = "$src1 = $dst" in
4364 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4366 } // ExeDomain = SSEPackedInt
4368 //===---------------------------------------------------------------------===//
4369 // SSE2 - Packed Mask Creation
4370 //===---------------------------------------------------------------------===//
4372 let ExeDomain = SSEPackedInt in {
4374 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4375 "pmovmskb\t{$src, $dst|$dst, $src}",
4376 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4377 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4378 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4380 let Predicates = [HasAVX2] in {
4381 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4382 "pmovmskb\t{$src, $dst|$dst, $src}",
4383 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4384 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4385 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4388 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4389 "pmovmskb\t{$src, $dst|$dst, $src}",
4390 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4392 } // ExeDomain = SSEPackedInt
4394 //===---------------------------------------------------------------------===//
4395 // SSE2 - Conditional Store
4396 //===---------------------------------------------------------------------===//
4398 let ExeDomain = SSEPackedInt in {
4401 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4402 (ins VR128:$src, VR128:$mask),
4403 "maskmovdqu\t{$mask, $src|$src, $mask}",
4404 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4406 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4407 (ins VR128:$src, VR128:$mask),
4408 "maskmovdqu\t{$mask, $src|$src, $mask}",
4409 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4412 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4413 "maskmovdqu\t{$mask, $src|$src, $mask}",
4414 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4416 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4417 "maskmovdqu\t{$mask, $src|$src, $mask}",
4418 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4420 } // ExeDomain = SSEPackedInt
4422 //===---------------------------------------------------------------------===//
4423 // SSE2 - Move Doubleword
4424 //===---------------------------------------------------------------------===//
4426 //===---------------------------------------------------------------------===//
4427 // Move Int Doubleword to Packed Double Int
4429 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4430 "movd\t{$src, $dst|$dst, $src}",
4432 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4433 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4436 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4438 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4439 "mov{d|q}\t{$src, $dst|$dst, $src}",
4441 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4442 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4443 "mov{d|q}\t{$src, $dst|$dst, $src}",
4444 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4446 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4447 "movd\t{$src, $dst|$dst, $src}",
4449 (v4i32 (scalar_to_vector GR32:$src)))]>;
4450 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4451 "movd\t{$src, $dst|$dst, $src}",
4453 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4454 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4455 "mov{d|q}\t{$src, $dst|$dst, $src}",
4457 (v2i64 (scalar_to_vector GR64:$src)))]>;
4458 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4459 "mov{d|q}\t{$src, $dst|$dst, $src}",
4460 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4462 //===---------------------------------------------------------------------===//
4463 // Move Int Doubleword to Single Scalar
4465 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4466 "movd\t{$src, $dst|$dst, $src}",
4467 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4469 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4470 "movd\t{$src, $dst|$dst, $src}",
4471 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4473 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4474 "movd\t{$src, $dst|$dst, $src}",
4475 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4477 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4478 "movd\t{$src, $dst|$dst, $src}",
4479 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4481 //===---------------------------------------------------------------------===//
4482 // Move Packed Doubleword Int to Packed Double Int
4484 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4485 "movd\t{$src, $dst|$dst, $src}",
4486 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4488 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4489 (ins i32mem:$dst, VR128:$src),
4490 "movd\t{$src, $dst|$dst, $src}",
4491 [(store (i32 (vector_extract (v4i32 VR128:$src),
4492 (iPTR 0))), addr:$dst)]>, VEX;
4493 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4494 "movd\t{$src, $dst|$dst, $src}",
4495 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4497 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4498 "movd\t{$src, $dst|$dst, $src}",
4499 [(store (i32 (vector_extract (v4i32 VR128:$src),
4500 (iPTR 0))), addr:$dst)]>;
4502 //===---------------------------------------------------------------------===//
4503 // Move Packed Doubleword Int first element to Doubleword Int
4505 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4506 "mov{d|q}\t{$src, $dst|$dst, $src}",
4507 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4509 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4511 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4512 "mov{d|q}\t{$src, $dst|$dst, $src}",
4513 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4516 //===---------------------------------------------------------------------===//
4517 // Bitcast FR64 <-> GR64
4519 let Predicates = [HasAVX] in
4520 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4521 "vmovq\t{$src, $dst|$dst, $src}",
4522 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4524 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4525 "mov{d|q}\t{$src, $dst|$dst, $src}",
4526 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4527 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4528 "movq\t{$src, $dst|$dst, $src}",
4529 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4531 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4532 "movq\t{$src, $dst|$dst, $src}",
4533 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4534 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4535 "mov{d|q}\t{$src, $dst|$dst, $src}",
4536 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4537 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4538 "movq\t{$src, $dst|$dst, $src}",
4539 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4541 //===---------------------------------------------------------------------===//
4542 // Move Scalar Single to Double Int
4544 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4545 "movd\t{$src, $dst|$dst, $src}",
4546 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4547 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4548 "movd\t{$src, $dst|$dst, $src}",
4549 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4550 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4551 "movd\t{$src, $dst|$dst, $src}",
4552 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4553 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4554 "movd\t{$src, $dst|$dst, $src}",
4555 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4557 //===---------------------------------------------------------------------===//
4558 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4560 let AddedComplexity = 15 in {
4561 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4562 "movd\t{$src, $dst|$dst, $src}",
4563 [(set VR128:$dst, (v4i32 (X86vzmovl
4564 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4566 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4567 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4568 [(set VR128:$dst, (v2i64 (X86vzmovl
4569 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4572 let AddedComplexity = 15 in {
4573 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4574 "movd\t{$src, $dst|$dst, $src}",
4575 [(set VR128:$dst, (v4i32 (X86vzmovl
4576 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4577 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4578 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4579 [(set VR128:$dst, (v2i64 (X86vzmovl
4580 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4583 let AddedComplexity = 20 in {
4584 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4585 "movd\t{$src, $dst|$dst, $src}",
4587 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4588 (loadi32 addr:$src))))))]>,
4590 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4593 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4594 (loadi32 addr:$src))))))]>;
4597 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4598 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4599 (MOVZDI2PDIrm addr:$src)>;
4600 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4601 (MOVZDI2PDIrm addr:$src)>;
4602 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4603 (MOVZDI2PDIrm addr:$src)>;
4606 let Predicates = [HasAVX] in {
4607 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4608 let AddedComplexity = 20 in {
4609 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4610 (VMOVZDI2PDIrm addr:$src)>;
4611 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4612 (VMOVZDI2PDIrm addr:$src)>;
4613 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4614 (VMOVZDI2PDIrm addr:$src)>;
4616 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4617 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4618 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4619 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4620 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4621 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4622 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4625 // These are the correct encodings of the instructions so that we know how to
4626 // read correct assembly, even though we continue to emit the wrong ones for
4627 // compatibility with Darwin's buggy assembler.
4628 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4629 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4630 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4631 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4632 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4633 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4634 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4635 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4636 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4637 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4638 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4639 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4641 //===---------------------------------------------------------------------===//
4642 // SSE2 - Move Quadword
4643 //===---------------------------------------------------------------------===//
4645 //===---------------------------------------------------------------------===//
4646 // Move Quadword Int to Packed Quadword Int
4648 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4649 "vmovq\t{$src, $dst|$dst, $src}",
4651 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4652 VEX, Requires<[HasAVX]>;
4653 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4654 "movq\t{$src, $dst|$dst, $src}",
4656 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4657 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4659 //===---------------------------------------------------------------------===//
4660 // Move Packed Quadword Int to Quadword Int
4662 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4663 "movq\t{$src, $dst|$dst, $src}",
4664 [(store (i64 (vector_extract (v2i64 VR128:$src),
4665 (iPTR 0))), addr:$dst)]>, VEX;
4666 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4667 "movq\t{$src, $dst|$dst, $src}",
4668 [(store (i64 (vector_extract (v2i64 VR128:$src),
4669 (iPTR 0))), addr:$dst)]>;
4671 //===---------------------------------------------------------------------===//
4672 // Store / copy lower 64-bits of a XMM register.
4674 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4675 "movq\t{$src, $dst|$dst, $src}",
4676 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4677 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4678 "movq\t{$src, $dst|$dst, $src}",
4679 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4681 let AddedComplexity = 20 in
4682 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4683 "vmovq\t{$src, $dst|$dst, $src}",
4685 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4686 (loadi64 addr:$src))))))]>,
4687 XS, VEX, Requires<[HasAVX]>;
4689 let AddedComplexity = 20 in
4690 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4691 "movq\t{$src, $dst|$dst, $src}",
4693 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4694 (loadi64 addr:$src))))))]>,
4695 XS, Requires<[HasSSE2]>;
4697 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4698 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4699 (MOVZQI2PQIrm addr:$src)>;
4700 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4701 (MOVZQI2PQIrm addr:$src)>;
4702 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4705 let Predicates = [HasAVX], AddedComplexity = 20 in {
4706 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4707 (VMOVZQI2PQIrm addr:$src)>;
4708 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4709 (VMOVZQI2PQIrm addr:$src)>;
4710 def : Pat<(v2i64 (X86vzload addr:$src)),
4711 (VMOVZQI2PQIrm addr:$src)>;
4714 //===---------------------------------------------------------------------===//
4715 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4716 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4718 let AddedComplexity = 15 in
4719 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4720 "vmovq\t{$src, $dst|$dst, $src}",
4721 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4722 XS, VEX, Requires<[HasAVX]>;
4723 let AddedComplexity = 15 in
4724 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4725 "movq\t{$src, $dst|$dst, $src}",
4726 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4727 XS, Requires<[HasSSE2]>;
4729 let AddedComplexity = 20 in
4730 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4731 "vmovq\t{$src, $dst|$dst, $src}",
4732 [(set VR128:$dst, (v2i64 (X86vzmovl
4733 (loadv2i64 addr:$src))))]>,
4734 XS, VEX, Requires<[HasAVX]>;
4735 let AddedComplexity = 20 in {
4736 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4737 "movq\t{$src, $dst|$dst, $src}",
4738 [(set VR128:$dst, (v2i64 (X86vzmovl
4739 (loadv2i64 addr:$src))))]>,
4740 XS, Requires<[HasSSE2]>;
4743 let AddedComplexity = 20 in {
4744 let Predicates = [HasSSE2] in {
4745 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4746 (MOVZPQILo2PQIrm addr:$src)>;
4747 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4748 (MOVZPQILo2PQIrr VR128:$src)>;
4750 let Predicates = [HasAVX] in {
4751 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4752 (VMOVZPQILo2PQIrm addr:$src)>;
4753 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4754 (VMOVZPQILo2PQIrr VR128:$src)>;
4758 // Instructions to match in the assembler
4759 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4760 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4761 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4762 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4763 // Recognize "movd" with GR64 destination, but encode as a "movq"
4764 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4765 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4767 // Instructions for the disassembler
4768 // xr = XMM register
4771 let Predicates = [HasAVX] in
4772 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4773 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4774 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4775 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4777 //===---------------------------------------------------------------------===//
4778 // SSE3 - Conversion Instructions
4779 //===---------------------------------------------------------------------===//
4781 // Convert Packed Double FP to Packed DW Integers
4782 let Predicates = [HasAVX] in {
4783 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4784 // register, but the same isn't true when using memory operands instead.
4785 // Provide other assembly rr and rm forms to address this explicitly.
4786 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4787 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4788 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4789 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4792 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4793 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4794 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4795 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4798 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4799 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4800 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4801 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4804 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4805 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4806 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4807 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4809 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4810 (VCVTPD2DQYrr VR256:$src)>;
4811 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4812 (VCVTPD2DQYrm addr:$src)>;
4814 // Convert Packed DW Integers to Packed Double FP
4815 let Predicates = [HasAVX] in {
4816 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4817 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4818 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4819 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4820 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4821 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4822 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4823 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4826 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4827 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4828 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4829 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4831 // AVX 256-bit register conversion intrinsics
4832 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4833 (VCVTDQ2PDYrr VR128:$src)>;
4834 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4835 (VCVTDQ2PDYrm addr:$src)>;
4837 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4838 (VCVTPD2DQYrr VR256:$src)>;
4839 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4840 (VCVTPD2DQYrm addr:$src)>;
4842 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4843 (VCVTDQ2PDYrr VR128:$src)>;
4844 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4845 (VCVTDQ2PDYrm addr:$src)>;
4847 //===---------------------------------------------------------------------===//
4848 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4849 //===---------------------------------------------------------------------===//
4850 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4851 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4852 X86MemOperand x86memop> {
4853 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4854 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4855 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4856 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4857 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4858 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4861 let Predicates = [HasAVX] in {
4862 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4863 v4f32, VR128, memopv4f32, f128mem>, VEX;
4864 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4865 v4f32, VR128, memopv4f32, f128mem>, VEX;
4866 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4867 v8f32, VR256, memopv8f32, f256mem>, VEX;
4868 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4869 v8f32, VR256, memopv8f32, f256mem>, VEX;
4871 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4872 memopv4f32, f128mem>;
4873 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4874 memopv4f32, f128mem>;
4876 let Predicates = [HasSSE3] in {
4877 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4878 (MOVSHDUPrr VR128:$src)>;
4879 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4880 (MOVSHDUPrm addr:$src)>;
4881 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4882 (MOVSLDUPrr VR128:$src)>;
4883 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4884 (MOVSLDUPrm addr:$src)>;
4887 let Predicates = [HasAVX] in {
4888 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4889 (VMOVSHDUPrr VR128:$src)>;
4890 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4891 (VMOVSHDUPrm addr:$src)>;
4892 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4893 (VMOVSLDUPrr VR128:$src)>;
4894 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4895 (VMOVSLDUPrm addr:$src)>;
4896 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4897 (VMOVSHDUPYrr VR256:$src)>;
4898 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4899 (VMOVSHDUPYrm addr:$src)>;
4900 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4901 (VMOVSLDUPYrr VR256:$src)>;
4902 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4903 (VMOVSLDUPYrm addr:$src)>;
4906 //===---------------------------------------------------------------------===//
4907 // SSE3 - Replicate Double FP - MOVDDUP
4908 //===---------------------------------------------------------------------===//
4910 multiclass sse3_replicate_dfp<string OpcodeStr> {
4911 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4912 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4913 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4914 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4915 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4917 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4921 // FIXME: Merge with above classe when there're patterns for the ymm version
4922 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4923 let Predicates = [HasAVX] in {
4924 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4925 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4927 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4928 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4933 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4934 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4935 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4937 let Predicates = [HasSSE3] in {
4938 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4940 (MOVDDUPrm addr:$src)>;
4941 let AddedComplexity = 5 in {
4942 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4943 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4944 (MOVDDUPrm addr:$src)>;
4945 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4946 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4947 (MOVDDUPrm addr:$src)>;
4949 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4950 (MOVDDUPrm addr:$src)>;
4951 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4952 (MOVDDUPrm addr:$src)>;
4953 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4954 (MOVDDUPrm addr:$src)>;
4955 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4956 (MOVDDUPrm addr:$src)>;
4957 def : Pat<(X86Movddup (bc_v2f64
4958 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4959 (MOVDDUPrm addr:$src)>;
4962 let Predicates = [HasAVX] in {
4963 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4965 (VMOVDDUPrm addr:$src)>;
4966 let AddedComplexity = 5 in {
4967 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4968 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4969 (VMOVDDUPrm addr:$src)>;
4970 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4971 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4972 (VMOVDDUPrm addr:$src)>;
4974 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4975 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4976 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4977 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4978 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4979 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4980 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4981 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4982 def : Pat<(X86Movddup (bc_v2f64
4983 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4984 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4987 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4988 (VMOVDDUPYrm addr:$src)>;
4989 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4990 (VMOVDDUPYrm addr:$src)>;
4991 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4992 (VMOVDDUPYrm addr:$src)>;
4993 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4994 (VMOVDDUPYrm addr:$src)>;
4995 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4996 (VMOVDDUPYrr VR256:$src)>;
4997 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4998 (VMOVDDUPYrr VR256:$src)>;
5001 //===---------------------------------------------------------------------===//
5002 // SSE3 - Move Unaligned Integer
5003 //===---------------------------------------------------------------------===//
5005 let Predicates = [HasAVX] in {
5006 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5007 "vlddqu\t{$src, $dst|$dst, $src}",
5008 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5009 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5010 "vlddqu\t{$src, $dst|$dst, $src}",
5011 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5013 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5014 "lddqu\t{$src, $dst|$dst, $src}",
5015 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5017 //===---------------------------------------------------------------------===//
5018 // SSE3 - Arithmetic
5019 //===---------------------------------------------------------------------===//
5021 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5022 X86MemOperand x86memop, bit Is2Addr = 1> {
5023 def rr : I<0xD0, MRMSrcReg,
5024 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5026 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5027 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5028 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5029 def rm : I<0xD0, MRMSrcMem,
5030 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5032 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5033 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5034 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5037 let Predicates = [HasAVX] in {
5038 let ExeDomain = SSEPackedSingle in {
5039 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5040 f128mem, 0>, TB, XD, VEX_4V;
5041 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5042 f256mem, 0>, TB, XD, VEX_4V;
5044 let ExeDomain = SSEPackedDouble in {
5045 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5046 f128mem, 0>, TB, OpSize, VEX_4V;
5047 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5048 f256mem, 0>, TB, OpSize, VEX_4V;
5051 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5052 let ExeDomain = SSEPackedSingle in
5053 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5055 let ExeDomain = SSEPackedDouble in
5056 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5057 f128mem>, TB, OpSize;
5060 //===---------------------------------------------------------------------===//
5061 // SSE3 Instructions
5062 //===---------------------------------------------------------------------===//
5065 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5066 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5067 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5069 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5070 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5071 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5073 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5075 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5076 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5077 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5079 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5080 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5081 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5083 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5084 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5085 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5087 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5089 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5091 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5094 let Predicates = [HasAVX] in {
5095 let ExeDomain = SSEPackedSingle in {
5096 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5097 X86fhadd, 0>, VEX_4V;
5098 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5099 X86fhsub, 0>, VEX_4V;
5100 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5101 X86fhadd, 0>, VEX_4V;
5102 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5103 X86fhsub, 0>, VEX_4V;
5105 let ExeDomain = SSEPackedDouble in {
5106 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5107 X86fhadd, 0>, VEX_4V;
5108 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5109 X86fhsub, 0>, VEX_4V;
5110 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5111 X86fhadd, 0>, VEX_4V;
5112 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5113 X86fhsub, 0>, VEX_4V;
5117 let Constraints = "$src1 = $dst" in {
5118 let ExeDomain = SSEPackedSingle in {
5119 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5120 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5122 let ExeDomain = SSEPackedDouble in {
5123 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5124 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5128 //===---------------------------------------------------------------------===//
5129 // SSSE3 - Packed Absolute Instructions
5130 //===---------------------------------------------------------------------===//
5133 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5134 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5135 Intrinsic IntId128> {
5136 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5138 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5139 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5142 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5144 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5147 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5150 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5151 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5152 Intrinsic IntId256> {
5153 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5155 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5156 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5159 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5161 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5164 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5167 let Predicates = [HasAVX] in {
5168 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5169 int_x86_ssse3_pabs_b_128>, VEX;
5170 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5171 int_x86_ssse3_pabs_w_128>, VEX;
5172 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5173 int_x86_ssse3_pabs_d_128>, VEX;
5176 let Predicates = [HasAVX2] in {
5177 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5178 int_x86_avx2_pabs_b>, VEX;
5179 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5180 int_x86_avx2_pabs_w>, VEX;
5181 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5182 int_x86_avx2_pabs_d>, VEX;
5185 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5186 int_x86_ssse3_pabs_b_128>;
5187 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5188 int_x86_ssse3_pabs_w_128>;
5189 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5190 int_x86_ssse3_pabs_d_128>;
5192 //===---------------------------------------------------------------------===//
5193 // SSSE3 - Packed Binary Operator Instructions
5194 //===---------------------------------------------------------------------===//
5196 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5197 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5198 Intrinsic IntId128, bit Is2Addr = 1> {
5199 let isCommutable = 1 in
5200 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5201 (ins VR128:$src1, VR128:$src2),
5203 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5204 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5205 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5207 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5208 (ins VR128:$src1, i128mem:$src2),
5210 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5211 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5213 (IntId128 VR128:$src1,
5214 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5217 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5218 Intrinsic IntId256> {
5219 let isCommutable = 1 in
5220 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5221 (ins VR256:$src1, VR256:$src2),
5222 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5223 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5225 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5226 (ins VR256:$src1, i256mem:$src2),
5227 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5229 (IntId256 VR256:$src1,
5230 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5233 let ImmT = NoImm, Predicates = [HasAVX] in {
5234 let isCommutable = 0 in {
5235 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5236 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5237 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5238 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5239 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5240 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5241 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5242 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5243 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5244 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5245 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5246 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5247 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5248 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5249 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5250 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5251 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5252 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5253 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5254 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5255 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5256 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5258 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5259 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5262 let ImmT = NoImm, Predicates = [HasAVX2] in {
5263 let isCommutable = 0 in {
5264 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5265 int_x86_avx2_phadd_w>, VEX_4V;
5266 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5267 int_x86_avx2_phadd_d>, VEX_4V;
5268 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5269 int_x86_avx2_phadd_sw>, VEX_4V;
5270 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5271 int_x86_avx2_phsub_w>, VEX_4V;
5272 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5273 int_x86_avx2_phsub_d>, VEX_4V;
5274 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5275 int_x86_avx2_phsub_sw>, VEX_4V;
5276 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5277 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5278 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5279 int_x86_avx2_pshuf_b>, VEX_4V;
5280 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5281 int_x86_avx2_psign_b>, VEX_4V;
5282 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5283 int_x86_avx2_psign_w>, VEX_4V;
5284 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5285 int_x86_avx2_psign_d>, VEX_4V;
5287 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5288 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5291 // None of these have i8 immediate fields.
5292 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5293 let isCommutable = 0 in {
5294 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5295 int_x86_ssse3_phadd_w_128>;
5296 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5297 int_x86_ssse3_phadd_d_128>;
5298 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5299 int_x86_ssse3_phadd_sw_128>;
5300 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5301 int_x86_ssse3_phsub_w_128>;
5302 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5303 int_x86_ssse3_phsub_d_128>;
5304 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5305 int_x86_ssse3_phsub_sw_128>;
5306 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5307 int_x86_ssse3_pmadd_ub_sw_128>;
5308 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5309 int_x86_ssse3_pshuf_b_128>;
5310 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5311 int_x86_ssse3_psign_b_128>;
5312 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5313 int_x86_ssse3_psign_w_128>;
5314 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5315 int_x86_ssse3_psign_d_128>;
5317 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5318 int_x86_ssse3_pmul_hr_sw_128>;
5321 let Predicates = [HasSSSE3] in {
5322 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5323 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5324 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5325 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5327 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5328 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5329 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5330 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5331 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5332 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5334 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5335 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5336 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5337 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5338 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5339 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5340 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5341 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5344 let Predicates = [HasAVX] in {
5345 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5346 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5347 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5348 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5350 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5351 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5352 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5353 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5354 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5355 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5357 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5358 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5359 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5360 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5361 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5362 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5363 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5364 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5367 let Predicates = [HasAVX2] in {
5368 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5369 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5370 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5371 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5372 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5373 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5375 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5376 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5377 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5378 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5379 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5380 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5381 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5382 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5385 //===---------------------------------------------------------------------===//
5386 // SSSE3 - Packed Align Instruction Patterns
5387 //===---------------------------------------------------------------------===//
5389 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5390 let neverHasSideEffects = 1 in {
5391 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5392 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5394 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5396 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5399 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5400 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5402 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5404 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5409 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5410 let neverHasSideEffects = 1 in {
5411 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5412 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5414 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5417 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5418 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5420 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5425 let Predicates = [HasAVX] in
5426 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5427 let Predicates = [HasAVX2] in
5428 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5429 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5430 defm PALIGN : ssse3_palign<"palignr">;
5432 let Predicates = [HasSSSE3] in {
5433 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5434 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5435 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5436 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5437 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5438 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5439 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5440 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5443 let Predicates = [HasAVX] in {
5444 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5445 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5446 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5447 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5448 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5449 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5450 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5451 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5454 //===---------------------------------------------------------------------===//
5455 // SSSE3 - Thread synchronization
5456 //===---------------------------------------------------------------------===//
5458 let usesCustomInserter = 1 in {
5459 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5460 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5461 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5462 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5465 let Uses = [EAX, ECX, EDX] in
5466 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5467 Requires<[HasSSE3]>;
5468 let Uses = [ECX, EAX] in
5469 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5470 Requires<[HasSSE3]>;
5472 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5473 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5475 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5476 Requires<[In32BitMode]>;
5477 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5478 Requires<[In64BitMode]>;
5480 //===----------------------------------------------------------------------===//
5481 // SSE4.1 - Packed Move with Sign/Zero Extend
5482 //===----------------------------------------------------------------------===//
5484 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5485 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5486 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5487 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5489 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5492 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5496 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5498 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5499 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5500 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5502 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5503 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5504 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5507 let Predicates = [HasAVX] in {
5508 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5510 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5512 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5514 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5516 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5518 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5522 let Predicates = [HasAVX2] in {
5523 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5524 int_x86_avx2_pmovsxbw>, VEX;
5525 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5526 int_x86_avx2_pmovsxwd>, VEX;
5527 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5528 int_x86_avx2_pmovsxdq>, VEX;
5529 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5530 int_x86_avx2_pmovzxbw>, VEX;
5531 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5532 int_x86_avx2_pmovzxwd>, VEX;
5533 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5534 int_x86_avx2_pmovzxdq>, VEX;
5537 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5538 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5539 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5540 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5541 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5542 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5544 let Predicates = [HasSSE41] in {
5545 // Common patterns involving scalar load.
5546 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5547 (PMOVSXBWrm addr:$src)>;
5548 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5549 (PMOVSXBWrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5552 (PMOVSXWDrm addr:$src)>;
5553 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5554 (PMOVSXWDrm addr:$src)>;
5556 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5557 (PMOVSXDQrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5559 (PMOVSXDQrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5562 (PMOVZXBWrm addr:$src)>;
5563 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5564 (PMOVZXBWrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5567 (PMOVZXWDrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5569 (PMOVZXWDrm addr:$src)>;
5571 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5572 (PMOVZXDQrm addr:$src)>;
5573 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5574 (PMOVZXDQrm addr:$src)>;
5577 let Predicates = [HasAVX] in {
5578 // Common patterns involving scalar load.
5579 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5580 (VPMOVSXBWrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5582 (VPMOVSXBWrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5585 (VPMOVSXWDrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5587 (VPMOVSXWDrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5590 (VPMOVSXDQrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5592 (VPMOVSXDQrm addr:$src)>;
5594 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5595 (VPMOVZXBWrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5597 (VPMOVZXBWrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5600 (VPMOVZXWDrm addr:$src)>;
5601 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5602 (VPMOVZXWDrm addr:$src)>;
5604 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5605 (VPMOVZXDQrm addr:$src)>;
5606 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5607 (VPMOVZXDQrm addr:$src)>;
5611 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5612 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5613 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5614 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5616 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5617 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5619 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5623 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5625 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5626 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5627 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5629 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5630 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5632 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5636 let Predicates = [HasAVX] in {
5637 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5639 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5641 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5643 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5647 let Predicates = [HasAVX2] in {
5648 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5649 int_x86_avx2_pmovsxbd>, VEX;
5650 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5651 int_x86_avx2_pmovsxwq>, VEX;
5652 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5653 int_x86_avx2_pmovzxbd>, VEX;
5654 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5655 int_x86_avx2_pmovzxwq>, VEX;
5658 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5659 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5660 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5661 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5663 let Predicates = [HasSSE41] in {
5664 // Common patterns involving scalar load
5665 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5666 (PMOVSXBDrm addr:$src)>;
5667 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5668 (PMOVSXWQrm addr:$src)>;
5670 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5671 (PMOVZXBDrm addr:$src)>;
5672 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5673 (PMOVZXWQrm addr:$src)>;
5676 let Predicates = [HasAVX] in {
5677 // Common patterns involving scalar load
5678 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5679 (VPMOVSXBDrm addr:$src)>;
5680 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5681 (VPMOVSXWQrm addr:$src)>;
5683 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5684 (VPMOVZXBDrm addr:$src)>;
5685 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5686 (VPMOVZXWQrm addr:$src)>;
5689 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5690 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5691 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5692 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5694 // Expecting a i16 load any extended to i32 value.
5695 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5696 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5697 [(set VR128:$dst, (IntId (bitconvert
5698 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5702 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5704 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5705 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5706 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5708 // Expecting a i16 load any extended to i32 value.
5709 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5710 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5711 [(set VR256:$dst, (IntId (bitconvert
5712 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5716 let Predicates = [HasAVX] in {
5717 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5719 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5722 let Predicates = [HasAVX2] in {
5723 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5724 int_x86_avx2_pmovsxbq>, VEX;
5725 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5726 int_x86_avx2_pmovzxbq>, VEX;
5728 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5729 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5731 let Predicates = [HasSSE41] in {
5732 // Common patterns involving scalar load
5733 def : Pat<(int_x86_sse41_pmovsxbq
5734 (bitconvert (v4i32 (X86vzmovl
5735 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5736 (PMOVSXBQrm addr:$src)>;
5738 def : Pat<(int_x86_sse41_pmovzxbq
5739 (bitconvert (v4i32 (X86vzmovl
5740 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5741 (PMOVZXBQrm addr:$src)>;
5744 let Predicates = [HasAVX] in {
5745 // Common patterns involving scalar load
5746 def : Pat<(int_x86_sse41_pmovsxbq
5747 (bitconvert (v4i32 (X86vzmovl
5748 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5749 (VPMOVSXBQrm addr:$src)>;
5751 def : Pat<(int_x86_sse41_pmovzxbq
5752 (bitconvert (v4i32 (X86vzmovl
5753 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5754 (VPMOVZXBQrm addr:$src)>;
5757 //===----------------------------------------------------------------------===//
5758 // SSE4.1 - Extract Instructions
5759 //===----------------------------------------------------------------------===//
5761 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5762 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5763 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5764 (ins VR128:$src1, i32i8imm:$src2),
5765 !strconcat(OpcodeStr,
5766 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5767 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5769 let neverHasSideEffects = 1, mayStore = 1 in
5770 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5771 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5772 !strconcat(OpcodeStr,
5773 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5776 // There's an AssertZext in the way of writing the store pattern
5777 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5780 let Predicates = [HasAVX] in {
5781 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5782 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5783 (ins VR128:$src1, i32i8imm:$src2),
5784 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5787 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5790 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5791 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5792 let neverHasSideEffects = 1, mayStore = 1 in
5793 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5794 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5795 !strconcat(OpcodeStr,
5796 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5799 // There's an AssertZext in the way of writing the store pattern
5800 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5803 let Predicates = [HasAVX] in
5804 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5806 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5809 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5810 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5811 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5812 (ins VR128:$src1, i32i8imm:$src2),
5813 !strconcat(OpcodeStr,
5814 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5816 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5817 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5818 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5819 !strconcat(OpcodeStr,
5820 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5821 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5822 addr:$dst)]>, OpSize;
5825 let Predicates = [HasAVX] in
5826 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5828 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5830 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5831 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5832 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5833 (ins VR128:$src1, i32i8imm:$src2),
5834 !strconcat(OpcodeStr,
5835 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5837 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5838 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5839 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5840 !strconcat(OpcodeStr,
5841 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5842 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5843 addr:$dst)]>, OpSize, REX_W;
5846 let Predicates = [HasAVX] in
5847 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5849 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5851 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5853 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5854 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5855 (ins VR128:$src1, i32i8imm:$src2),
5856 !strconcat(OpcodeStr,
5857 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5859 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5861 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5862 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5863 !strconcat(OpcodeStr,
5864 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5865 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5866 addr:$dst)]>, OpSize;
5869 let ExeDomain = SSEPackedSingle in {
5870 let Predicates = [HasAVX] in {
5871 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5872 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5873 (ins VR128:$src1, i32i8imm:$src2),
5874 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5877 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5880 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5881 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5884 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5885 Requires<[HasSSE41]>;
5886 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5889 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5892 //===----------------------------------------------------------------------===//
5893 // SSE4.1 - Insert Instructions
5894 //===----------------------------------------------------------------------===//
5896 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5897 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5898 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5900 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5902 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5904 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5905 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5906 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5908 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5910 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5912 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5913 imm:$src3))]>, OpSize;
5916 let Predicates = [HasAVX] in
5917 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5918 let Constraints = "$src1 = $dst" in
5919 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5921 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5922 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5923 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5925 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5927 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5929 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5931 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5932 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5934 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5936 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5938 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5939 imm:$src3)))]>, OpSize;
5942 let Predicates = [HasAVX] in
5943 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5944 let Constraints = "$src1 = $dst" in
5945 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5947 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5948 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5949 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5951 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5953 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5955 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5957 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5958 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5960 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5962 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5964 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5965 imm:$src3)))]>, OpSize;
5968 let Predicates = [HasAVX] in
5969 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5970 let Constraints = "$src1 = $dst" in
5971 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5973 // insertps has a few different modes, there's the first two here below which
5974 // are optimized inserts that won't zero arbitrary elements in the destination
5975 // vector. The next one matches the intrinsic and could zero arbitrary elements
5976 // in the target vector.
5977 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5978 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5979 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5981 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5983 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5985 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5987 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5988 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5990 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5992 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5994 (X86insrtps VR128:$src1,
5995 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5996 imm:$src3))]>, OpSize;
5999 let ExeDomain = SSEPackedSingle in {
6000 let Constraints = "$src1 = $dst" in
6001 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6002 let Predicates = [HasAVX] in
6003 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6006 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6007 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6009 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6010 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6011 Requires<[HasSSE41]>;
6013 //===----------------------------------------------------------------------===//
6014 // SSE4.1 - Round Instructions
6015 //===----------------------------------------------------------------------===//
6017 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6018 X86MemOperand x86memop, RegisterClass RC,
6019 PatFrag mem_frag32, PatFrag mem_frag64,
6020 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6021 let ExeDomain = SSEPackedSingle in {
6022 // Intrinsic operation, reg.
6023 // Vector intrinsic operation, reg
6024 def PSr : SS4AIi8<opcps, MRMSrcReg,
6025 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6026 !strconcat(OpcodeStr,
6027 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6028 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6031 // Vector intrinsic operation, mem
6032 def PSm : SS4AIi8<opcps, MRMSrcMem,
6033 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6034 !strconcat(OpcodeStr,
6035 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6037 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6039 } // ExeDomain = SSEPackedSingle
6041 let ExeDomain = SSEPackedDouble in {
6042 // Vector intrinsic operation, reg
6043 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6044 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6045 !strconcat(OpcodeStr,
6046 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6047 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6050 // Vector intrinsic operation, mem
6051 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6052 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6053 !strconcat(OpcodeStr,
6054 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6056 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6058 } // ExeDomain = SSEPackedDouble
6061 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6064 Intrinsic F64Int, bit Is2Addr = 1> {
6065 let ExeDomain = GenericDomain in {
6067 def SSr : SS4AIi8<opcss, MRMSrcReg,
6068 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6070 !strconcat(OpcodeStr,
6071 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6072 !strconcat(OpcodeStr,
6073 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6076 // Intrinsic operation, reg.
6077 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6078 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6080 !strconcat(OpcodeStr,
6081 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6082 !strconcat(OpcodeStr,
6083 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6084 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6087 // Intrinsic operation, mem.
6088 def SSm : SS4AIi8<opcss, MRMSrcMem,
6089 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6091 !strconcat(OpcodeStr,
6092 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6093 !strconcat(OpcodeStr,
6094 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6096 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6100 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6101 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6103 !strconcat(OpcodeStr,
6104 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6105 !strconcat(OpcodeStr,
6106 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6109 // Intrinsic operation, reg.
6110 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6111 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6113 !strconcat(OpcodeStr,
6114 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6115 !strconcat(OpcodeStr,
6116 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6117 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6120 // Intrinsic operation, mem.
6121 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6122 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6124 !strconcat(OpcodeStr,
6125 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6126 !strconcat(OpcodeStr,
6127 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6129 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6131 } // ExeDomain = GenericDomain
6134 // FP round - roundss, roundps, roundsd, roundpd
6135 let Predicates = [HasAVX] in {
6137 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6138 memopv4f32, memopv2f64,
6139 int_x86_sse41_round_ps,
6140 int_x86_sse41_round_pd>, VEX;
6141 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6142 memopv8f32, memopv4f64,
6143 int_x86_avx_round_ps_256,
6144 int_x86_avx_round_pd_256>, VEX;
6145 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6146 int_x86_sse41_round_ss,
6147 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6149 def : Pat<(ffloor FR32:$src),
6150 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6151 def : Pat<(f64 (ffloor FR64:$src)),
6152 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6153 def : Pat<(f32 (fnearbyint FR32:$src)),
6154 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6155 def : Pat<(f64 (fnearbyint FR64:$src)),
6156 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6157 def : Pat<(f32 (fceil FR32:$src)),
6158 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6159 def : Pat<(f64 (fceil FR64:$src)),
6160 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6161 def : Pat<(f32 (frint FR32:$src)),
6162 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6163 def : Pat<(f64 (frint FR64:$src)),
6164 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6165 def : Pat<(f32 (ftrunc FR32:$src)),
6166 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6167 def : Pat<(f64 (ftrunc FR64:$src)),
6168 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6171 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6172 memopv4f32, memopv2f64,
6173 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6174 let Constraints = "$src1 = $dst" in
6175 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6176 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6178 def : Pat<(ffloor FR32:$src),
6179 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6180 def : Pat<(f64 (ffloor FR64:$src)),
6181 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6182 def : Pat<(f32 (fnearbyint FR32:$src)),
6183 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6184 def : Pat<(f64 (fnearbyint FR64:$src)),
6185 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6186 def : Pat<(f32 (fceil FR32:$src)),
6187 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6188 def : Pat<(f64 (fceil FR64:$src)),
6189 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6190 def : Pat<(f32 (frint FR32:$src)),
6191 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6192 def : Pat<(f64 (frint FR64:$src)),
6193 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6194 def : Pat<(f32 (ftrunc FR32:$src)),
6195 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6196 def : Pat<(f64 (ftrunc FR64:$src)),
6197 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6199 //===----------------------------------------------------------------------===//
6200 // SSE4.1 - Packed Bit Test
6201 //===----------------------------------------------------------------------===//
6203 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6204 // the intel intrinsic that corresponds to this.
6205 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6206 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6207 "vptest\t{$src2, $src1|$src1, $src2}",
6208 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6210 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6211 "vptest\t{$src2, $src1|$src1, $src2}",
6212 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6215 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6216 "vptest\t{$src2, $src1|$src1, $src2}",
6217 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6219 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6220 "vptest\t{$src2, $src1|$src1, $src2}",
6221 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6225 let Defs = [EFLAGS] in {
6226 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6227 "ptest\t{$src2, $src1|$src1, $src2}",
6228 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6230 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6231 "ptest\t{$src2, $src1|$src1, $src2}",
6232 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6236 // The bit test instructions below are AVX only
6237 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6238 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6239 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6240 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6241 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6242 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6243 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6244 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6248 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6249 let ExeDomain = SSEPackedSingle in {
6250 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6251 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6253 let ExeDomain = SSEPackedDouble in {
6254 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6255 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6259 //===----------------------------------------------------------------------===//
6260 // SSE4.1 - Misc Instructions
6261 //===----------------------------------------------------------------------===//
6263 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6264 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6265 "popcnt{w}\t{$src, $dst|$dst, $src}",
6266 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6268 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6269 "popcnt{w}\t{$src, $dst|$dst, $src}",
6270 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6271 (implicit EFLAGS)]>, OpSize, XS;
6273 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6274 "popcnt{l}\t{$src, $dst|$dst, $src}",
6275 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6277 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6278 "popcnt{l}\t{$src, $dst|$dst, $src}",
6279 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6280 (implicit EFLAGS)]>, XS;
6282 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6283 "popcnt{q}\t{$src, $dst|$dst, $src}",
6284 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6286 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6287 "popcnt{q}\t{$src, $dst|$dst, $src}",
6288 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6289 (implicit EFLAGS)]>, XS;
6294 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6295 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6296 Intrinsic IntId128> {
6297 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6299 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6300 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6301 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6303 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6306 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6309 let Predicates = [HasAVX] in
6310 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6311 int_x86_sse41_phminposuw>, VEX;
6312 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6313 int_x86_sse41_phminposuw>;
6315 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6316 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6317 Intrinsic IntId128, bit Is2Addr = 1> {
6318 let isCommutable = 1 in
6319 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6320 (ins VR128:$src1, VR128:$src2),
6322 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6323 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6324 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6325 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6326 (ins VR128:$src1, i128mem:$src2),
6328 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6331 (IntId128 VR128:$src1,
6332 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6335 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6336 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6337 Intrinsic IntId256> {
6338 let isCommutable = 1 in
6339 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6340 (ins VR256:$src1, VR256:$src2),
6341 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6342 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6343 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6344 (ins VR256:$src1, i256mem:$src2),
6345 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6347 (IntId256 VR256:$src1,
6348 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6351 let Predicates = [HasAVX] in {
6352 let isCommutable = 0 in
6353 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6355 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6357 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6359 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6361 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6363 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6365 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6367 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6369 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6371 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6373 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6376 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6377 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6378 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6379 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6382 let Predicates = [HasAVX2] in {
6383 let isCommutable = 0 in
6384 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6385 int_x86_avx2_packusdw>, VEX_4V;
6386 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6387 int_x86_avx2_pcmpeq_q>, VEX_4V;
6388 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6389 int_x86_avx2_pmins_b>, VEX_4V;
6390 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6391 int_x86_avx2_pmins_d>, VEX_4V;
6392 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6393 int_x86_avx2_pminu_d>, VEX_4V;
6394 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6395 int_x86_avx2_pminu_w>, VEX_4V;
6396 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6397 int_x86_avx2_pmaxs_b>, VEX_4V;
6398 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6399 int_x86_avx2_pmaxs_d>, VEX_4V;
6400 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6401 int_x86_avx2_pmaxu_d>, VEX_4V;
6402 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6403 int_x86_avx2_pmaxu_w>, VEX_4V;
6404 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6405 int_x86_avx2_pmul_dq>, VEX_4V;
6407 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6408 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6409 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6410 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6413 let Constraints = "$src1 = $dst" in {
6414 let isCommutable = 0 in
6415 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6416 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6417 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6418 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6419 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6420 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6421 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6422 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6423 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6424 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6425 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6428 let Predicates = [HasSSE41] in {
6429 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6430 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6431 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6432 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6435 /// SS48I_binop_rm - Simple SSE41 binary operator.
6436 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6437 ValueType OpVT, bit Is2Addr = 1> {
6438 let isCommutable = 1 in
6439 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6440 (ins VR128:$src1, VR128:$src2),
6442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6444 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6446 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6447 (ins VR128:$src1, i128mem:$src2),
6449 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6450 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6451 [(set VR128:$dst, (OpNode VR128:$src1,
6452 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6456 /// SS48I_binop_rm - Simple SSE41 binary operator.
6457 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6459 let isCommutable = 1 in
6460 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6461 (ins VR256:$src1, VR256:$src2),
6462 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6463 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6465 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6466 (ins VR256:$src1, i256mem:$src2),
6467 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6468 [(set VR256:$dst, (OpNode VR256:$src1,
6469 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6473 let Predicates = [HasAVX] in
6474 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6475 let Predicates = [HasAVX2] in
6476 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6477 let Constraints = "$src1 = $dst" in
6478 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6480 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6481 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6482 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6483 X86MemOperand x86memop, bit Is2Addr = 1> {
6484 let isCommutable = 1 in
6485 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6486 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6488 !strconcat(OpcodeStr,
6489 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6490 !strconcat(OpcodeStr,
6491 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6492 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6494 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6495 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6497 !strconcat(OpcodeStr,
6498 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6499 !strconcat(OpcodeStr,
6500 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6503 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6507 let Predicates = [HasAVX] in {
6508 let isCommutable = 0 in {
6509 let ExeDomain = SSEPackedSingle in {
6510 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6511 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6512 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6513 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6515 let ExeDomain = SSEPackedDouble in {
6516 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6517 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6518 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6519 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6521 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6522 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6523 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6524 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6526 let ExeDomain = SSEPackedSingle in
6527 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6528 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6529 let ExeDomain = SSEPackedDouble in
6530 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6531 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6532 let ExeDomain = SSEPackedSingle in
6533 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6534 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6537 let Predicates = [HasAVX2] in {
6538 let isCommutable = 0 in {
6539 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6540 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6541 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6542 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6546 let Constraints = "$src1 = $dst" in {
6547 let isCommutable = 0 in {
6548 let ExeDomain = SSEPackedSingle in
6549 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6550 VR128, memopv4f32, i128mem>;
6551 let ExeDomain = SSEPackedDouble in
6552 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6553 VR128, memopv2f64, i128mem>;
6554 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6555 VR128, memopv2i64, i128mem>;
6556 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6557 VR128, memopv2i64, i128mem>;
6559 let ExeDomain = SSEPackedSingle in
6560 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6561 VR128, memopv4f32, i128mem>;
6562 let ExeDomain = SSEPackedDouble in
6563 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6564 VR128, memopv2f64, i128mem>;
6567 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6568 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6569 RegisterClass RC, X86MemOperand x86memop,
6570 PatFrag mem_frag, Intrinsic IntId> {
6571 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6572 (ins RC:$src1, RC:$src2, RC:$src3),
6573 !strconcat(OpcodeStr,
6574 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6575 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6576 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6578 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6579 (ins RC:$src1, x86memop:$src2, RC:$src3),
6580 !strconcat(OpcodeStr,
6581 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6583 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6585 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6588 let Predicates = [HasAVX] in {
6589 let ExeDomain = SSEPackedDouble in {
6590 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6591 memopv2f64, int_x86_sse41_blendvpd>;
6592 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6593 memopv4f64, int_x86_avx_blendv_pd_256>;
6594 } // ExeDomain = SSEPackedDouble
6595 let ExeDomain = SSEPackedSingle in {
6596 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6597 memopv4f32, int_x86_sse41_blendvps>;
6598 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6599 memopv8f32, int_x86_avx_blendv_ps_256>;
6600 } // ExeDomain = SSEPackedSingle
6601 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6602 memopv2i64, int_x86_sse41_pblendvb>;
6605 let Predicates = [HasAVX2] in {
6606 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6607 memopv4i64, int_x86_avx2_pblendvb>;
6610 let Predicates = [HasAVX] in {
6611 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6612 (v16i8 VR128:$src2))),
6613 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6614 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6615 (v4i32 VR128:$src2))),
6616 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6617 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6618 (v4f32 VR128:$src2))),
6619 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6620 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6621 (v2i64 VR128:$src2))),
6622 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6623 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6624 (v2f64 VR128:$src2))),
6625 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6626 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6627 (v8i32 VR256:$src2))),
6628 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6629 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6630 (v8f32 VR256:$src2))),
6631 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6632 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6633 (v4i64 VR256:$src2))),
6634 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6635 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6636 (v4f64 VR256:$src2))),
6637 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6640 let Predicates = [HasAVX2] in {
6641 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6642 (v32i8 VR256:$src2))),
6643 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6646 /// SS41I_ternary_int - SSE 4.1 ternary operator
6647 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6648 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6650 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6651 (ins VR128:$src1, VR128:$src2),
6652 !strconcat(OpcodeStr,
6653 "\t{$src2, $dst|$dst, $src2}"),
6654 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6657 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6658 (ins VR128:$src1, i128mem:$src2),
6659 !strconcat(OpcodeStr,
6660 "\t{$src2, $dst|$dst, $src2}"),
6663 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6667 let ExeDomain = SSEPackedDouble in
6668 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6669 int_x86_sse41_blendvpd>;
6670 let ExeDomain = SSEPackedSingle in
6671 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6672 int_x86_sse41_blendvps>;
6673 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6674 int_x86_sse41_pblendvb>;
6676 let Predicates = [HasSSE41] in {
6677 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6678 (v16i8 VR128:$src2))),
6679 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6680 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6681 (v4i32 VR128:$src2))),
6682 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6683 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6684 (v4f32 VR128:$src2))),
6685 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6686 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6687 (v2i64 VR128:$src2))),
6688 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6689 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6690 (v2f64 VR128:$src2))),
6691 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6694 let Predicates = [HasAVX] in
6695 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6696 "vmovntdqa\t{$src, $dst|$dst, $src}",
6697 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6699 let Predicates = [HasAVX2] in
6700 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6701 "vmovntdqa\t{$src, $dst|$dst, $src}",
6702 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6704 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6705 "movntdqa\t{$src, $dst|$dst, $src}",
6706 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6709 //===----------------------------------------------------------------------===//
6710 // SSE4.2 - Compare Instructions
6711 //===----------------------------------------------------------------------===//
6713 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6714 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6715 Intrinsic IntId128, bit Is2Addr = 1> {
6716 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6717 (ins VR128:$src1, VR128:$src2),
6719 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6720 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6721 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6723 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6724 (ins VR128:$src1, i128mem:$src2),
6726 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6727 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6729 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6732 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6733 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6734 Intrinsic IntId256> {
6735 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6736 (ins VR256:$src1, VR256:$src2),
6737 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6738 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6740 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6741 (ins VR256:$src1, i256mem:$src2),
6742 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6744 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6747 let Predicates = [HasAVX] in {
6748 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6751 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6752 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6753 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6754 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6757 let Predicates = [HasAVX2] in {
6758 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6761 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6762 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6763 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6764 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6767 let Constraints = "$src1 = $dst" in
6768 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6770 let Predicates = [HasSSE42] in {
6771 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6772 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6773 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6774 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6777 //===----------------------------------------------------------------------===//
6778 // SSE4.2 - String/text Processing Instructions
6779 //===----------------------------------------------------------------------===//
6781 // Packed Compare Implicit Length Strings, Return Mask
6782 multiclass pseudo_pcmpistrm<string asm> {
6783 def REG : PseudoI<(outs VR128:$dst),
6784 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6785 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6787 def MEM : PseudoI<(outs VR128:$dst),
6788 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6789 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6790 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6793 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6794 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6795 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6798 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6799 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6800 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6801 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6803 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6804 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6805 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6808 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6809 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6810 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6811 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6813 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6814 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6815 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6818 // Packed Compare Explicit Length Strings, Return Mask
6819 multiclass pseudo_pcmpestrm<string asm> {
6820 def REG : PseudoI<(outs VR128:$dst),
6821 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6822 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6823 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6824 def MEM : PseudoI<(outs VR128:$dst),
6825 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6826 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6827 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6830 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6831 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6832 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6835 let Predicates = [HasAVX],
6836 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6837 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6838 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6839 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6841 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6842 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6843 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6846 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6847 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6848 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6849 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6851 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6852 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6853 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6856 // Packed Compare Implicit Length Strings, Return Index
6857 let Defs = [ECX, EFLAGS] in {
6858 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6859 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6860 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6861 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6862 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6863 (implicit EFLAGS)]>, OpSize;
6864 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6865 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6866 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6867 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6868 (implicit EFLAGS)]>, OpSize;
6872 let Predicates = [HasAVX] in {
6873 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6875 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6877 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6879 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6881 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6883 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6887 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6888 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6889 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6890 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6891 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6892 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6894 // Packed Compare Explicit Length Strings, Return Index
6895 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6896 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6897 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6898 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6899 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6900 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6901 (implicit EFLAGS)]>, OpSize;
6902 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6903 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6904 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6906 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6907 (implicit EFLAGS)]>, OpSize;
6911 let Predicates = [HasAVX] in {
6912 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6914 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6916 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6918 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6920 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6922 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6926 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6927 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6928 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6929 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6930 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6931 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6933 //===----------------------------------------------------------------------===//
6934 // SSE4.2 - CRC Instructions
6935 //===----------------------------------------------------------------------===//
6937 // No CRC instructions have AVX equivalents
6939 // crc intrinsic instruction
6940 // This set of instructions are only rm, the only difference is the size
6942 let Constraints = "$src1 = $dst" in {
6943 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6944 (ins GR32:$src1, i8mem:$src2),
6945 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6947 (int_x86_sse42_crc32_32_8 GR32:$src1,
6948 (load addr:$src2)))]>;
6949 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6950 (ins GR32:$src1, GR8:$src2),
6951 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6953 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6954 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6955 (ins GR32:$src1, i16mem:$src2),
6956 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6958 (int_x86_sse42_crc32_32_16 GR32:$src1,
6959 (load addr:$src2)))]>,
6961 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6962 (ins GR32:$src1, GR16:$src2),
6963 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6965 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6967 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6968 (ins GR32:$src1, i32mem:$src2),
6969 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6971 (int_x86_sse42_crc32_32_32 GR32:$src1,
6972 (load addr:$src2)))]>;
6973 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6974 (ins GR32:$src1, GR32:$src2),
6975 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6977 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6978 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6979 (ins GR64:$src1, i8mem:$src2),
6980 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6982 (int_x86_sse42_crc32_64_8 GR64:$src1,
6983 (load addr:$src2)))]>,
6985 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6986 (ins GR64:$src1, GR8:$src2),
6987 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6989 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6991 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6992 (ins GR64:$src1, i64mem:$src2),
6993 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6995 (int_x86_sse42_crc32_64_64 GR64:$src1,
6996 (load addr:$src2)))]>,
6998 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6999 (ins GR64:$src1, GR64:$src2),
7000 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7002 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7006 //===----------------------------------------------------------------------===//
7007 // AES-NI Instructions
7008 //===----------------------------------------------------------------------===//
7010 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7011 Intrinsic IntId128, bit Is2Addr = 1> {
7012 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7013 (ins VR128:$src1, VR128:$src2),
7015 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7016 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7017 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7019 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7020 (ins VR128:$src1, i128mem:$src2),
7022 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7023 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7025 (IntId128 VR128:$src1,
7026 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
7029 // Perform One Round of an AES Encryption/Decryption Flow
7030 let Predicates = [HasAVX, HasAES] in {
7031 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7032 int_x86_aesni_aesenc, 0>, VEX_4V;
7033 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7034 int_x86_aesni_aesenclast, 0>, VEX_4V;
7035 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7036 int_x86_aesni_aesdec, 0>, VEX_4V;
7037 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7038 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7041 let Constraints = "$src1 = $dst" in {
7042 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7043 int_x86_aesni_aesenc>;
7044 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7045 int_x86_aesni_aesenclast>;
7046 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7047 int_x86_aesni_aesdec>;
7048 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7049 int_x86_aesni_aesdeclast>;
7052 let Predicates = [HasAES] in {
7053 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
7054 (AESENCrr VR128:$src1, VR128:$src2)>;
7055 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
7056 (AESENCrm VR128:$src1, addr:$src2)>;
7057 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7058 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
7059 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7060 (AESENCLASTrm VR128:$src1, addr:$src2)>;
7061 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7062 (AESDECrr VR128:$src1, VR128:$src2)>;
7063 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7064 (AESDECrm VR128:$src1, addr:$src2)>;
7065 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7066 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
7067 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7068 (AESDECLASTrm VR128:$src1, addr:$src2)>;
7071 let Predicates = [HasAVX, HasAES], AddedComplexity = 20 in {
7072 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
7073 (VAESENCrr VR128:$src1, VR128:$src2)>;
7074 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
7075 (VAESENCrm VR128:$src1, addr:$src2)>;
7076 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7077 (VAESENCLASTrr VR128:$src1, VR128:$src2)>;
7078 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7079 (VAESENCLASTrm VR128:$src1, addr:$src2)>;
7080 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7081 (VAESDECrr VR128:$src1, VR128:$src2)>;
7082 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7083 (VAESDECrm VR128:$src1, addr:$src2)>;
7084 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7085 (VAESDECLASTrr VR128:$src1, VR128:$src2)>;
7086 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7087 (VAESDECLASTrm VR128:$src1, addr:$src2)>;
7090 // Perform the AES InvMixColumn Transformation
7091 let Predicates = [HasAVX, HasAES] in {
7092 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7094 "vaesimc\t{$src1, $dst|$dst, $src1}",
7096 (int_x86_aesni_aesimc VR128:$src1))]>,
7098 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7099 (ins i128mem:$src1),
7100 "vaesimc\t{$src1, $dst|$dst, $src1}",
7102 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7105 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7107 "aesimc\t{$src1, $dst|$dst, $src1}",
7109 (int_x86_aesni_aesimc VR128:$src1))]>,
7111 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7112 (ins i128mem:$src1),
7113 "aesimc\t{$src1, $dst|$dst, $src1}",
7115 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7118 // AES Round Key Generation Assist
7119 let Predicates = [HasAVX, HasAES] in {
7120 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7121 (ins VR128:$src1, i8imm:$src2),
7122 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7124 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7126 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7127 (ins i128mem:$src1, i8imm:$src2),
7128 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7130 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7134 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7135 (ins VR128:$src1, i8imm:$src2),
7136 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7138 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7140 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7141 (ins i128mem:$src1, i8imm:$src2),
7142 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7144 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7148 //===----------------------------------------------------------------------===//
7149 // CLMUL Instructions
7150 //===----------------------------------------------------------------------===//
7152 // Carry-less Multiplication instructions
7153 let neverHasSideEffects = 1 in {
7154 let Constraints = "$src1 = $dst" in {
7155 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7156 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7157 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7161 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7162 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7163 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7167 // AVX carry-less Multiplication instructions
7168 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7169 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7170 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7174 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7175 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7176 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7181 multiclass pclmul_alias<string asm, int immop> {
7182 def : InstAlias<!strconcat("pclmul", asm,
7183 "dq {$src, $dst|$dst, $src}"),
7184 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7186 def : InstAlias<!strconcat("pclmul", asm,
7187 "dq {$src, $dst|$dst, $src}"),
7188 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7190 def : InstAlias<!strconcat("vpclmul", asm,
7191 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7192 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7194 def : InstAlias<!strconcat("vpclmul", asm,
7195 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7196 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7198 defm : pclmul_alias<"hqhq", 0x11>;
7199 defm : pclmul_alias<"hqlq", 0x01>;
7200 defm : pclmul_alias<"lqhq", 0x10>;
7201 defm : pclmul_alias<"lqlq", 0x00>;
7203 //===----------------------------------------------------------------------===//
7205 //===----------------------------------------------------------------------===//
7207 //===----------------------------------------------------------------------===//
7208 // VBROADCAST - Load from memory and broadcast to all elements of the
7209 // destination operand
7211 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7212 X86MemOperand x86memop, Intrinsic Int> :
7213 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7214 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7215 [(set RC:$dst, (Int addr:$src))]>, VEX;
7217 // AVX2 adds register forms
7218 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7220 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7221 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7222 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7224 let ExeDomain = SSEPackedSingle in {
7225 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7226 int_x86_avx_vbroadcast_ss>;
7227 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7228 int_x86_avx_vbroadcast_ss_256>;
7230 let ExeDomain = SSEPackedDouble in
7231 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7232 int_x86_avx_vbroadcast_sd_256>;
7233 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7234 int_x86_avx_vbroadcastf128_pd_256>;
7236 let ExeDomain = SSEPackedSingle in {
7237 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7238 int_x86_avx2_vbroadcast_ss_ps>;
7239 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7240 int_x86_avx2_vbroadcast_ss_ps_256>;
7242 let ExeDomain = SSEPackedDouble in
7243 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7244 int_x86_avx2_vbroadcast_sd_pd_256>;
7246 let Predicates = [HasAVX2] in
7247 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7248 int_x86_avx2_vbroadcasti128>;
7250 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7251 (VBROADCASTF128 addr:$src)>;
7254 //===----------------------------------------------------------------------===//
7255 // VINSERTF128 - Insert packed floating-point values
7257 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7258 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7259 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7260 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7263 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7264 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7265 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7269 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7270 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7271 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7272 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7273 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7274 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7276 //===----------------------------------------------------------------------===//
7277 // VEXTRACTF128 - Extract packed floating-point values
7279 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7280 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7281 (ins VR256:$src1, i8imm:$src2),
7282 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7285 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7286 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7287 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7291 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7292 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7293 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7294 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7295 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7296 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7298 //===----------------------------------------------------------------------===//
7299 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7301 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7302 Intrinsic IntLd, Intrinsic IntLd256,
7303 Intrinsic IntSt, Intrinsic IntSt256> {
7304 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7305 (ins VR128:$src1, f128mem:$src2),
7306 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7307 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7309 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7310 (ins VR256:$src1, f256mem:$src2),
7311 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7312 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7314 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7315 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7316 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7317 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7318 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7319 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7320 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7321 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7324 let ExeDomain = SSEPackedSingle in
7325 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7326 int_x86_avx_maskload_ps,
7327 int_x86_avx_maskload_ps_256,
7328 int_x86_avx_maskstore_ps,
7329 int_x86_avx_maskstore_ps_256>;
7330 let ExeDomain = SSEPackedDouble in
7331 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7332 int_x86_avx_maskload_pd,
7333 int_x86_avx_maskload_pd_256,
7334 int_x86_avx_maskstore_pd,
7335 int_x86_avx_maskstore_pd_256>;
7337 //===----------------------------------------------------------------------===//
7338 // VPERMIL - Permute Single and Double Floating-Point Values
7340 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7341 RegisterClass RC, X86MemOperand x86memop_f,
7342 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7343 Intrinsic IntVar, Intrinsic IntImm> {
7344 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7345 (ins RC:$src1, RC:$src2),
7346 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7347 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7348 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7349 (ins RC:$src1, x86memop_i:$src2),
7350 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7351 [(set RC:$dst, (IntVar RC:$src1,
7352 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7354 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7355 (ins RC:$src1, i8imm:$src2),
7356 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7357 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7358 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7359 (ins x86memop_f:$src1, i8imm:$src2),
7360 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7361 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7364 let ExeDomain = SSEPackedSingle in {
7365 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7366 memopv4f32, memopv2i64,
7367 int_x86_avx_vpermilvar_ps,
7368 int_x86_avx_vpermil_ps>;
7369 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7370 memopv8f32, memopv4i64,
7371 int_x86_avx_vpermilvar_ps_256,
7372 int_x86_avx_vpermil_ps_256>;
7374 let ExeDomain = SSEPackedDouble in {
7375 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7376 memopv2f64, memopv2i64,
7377 int_x86_avx_vpermilvar_pd,
7378 int_x86_avx_vpermil_pd>;
7379 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7380 memopv4f64, memopv4i64,
7381 int_x86_avx_vpermilvar_pd_256,
7382 int_x86_avx_vpermil_pd_256>;
7385 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7386 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7387 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7388 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7389 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7390 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7391 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7392 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7393 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7394 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7395 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7396 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7397 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7399 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7400 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7401 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7403 //===----------------------------------------------------------------------===//
7404 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7406 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7407 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7408 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7409 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7412 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7413 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7414 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7418 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7419 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7420 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7421 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7422 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7423 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7425 def : Pat<(int_x86_avx_vperm2f128_ps_256
7426 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7427 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7428 def : Pat<(int_x86_avx_vperm2f128_pd_256
7429 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7430 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7431 def : Pat<(int_x86_avx_vperm2f128_si_256
7432 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7433 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7435 //===----------------------------------------------------------------------===//
7436 // VZERO - Zero YMM registers
7438 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7439 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7440 // Zero All YMM registers
7441 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7442 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7444 // Zero Upper bits of YMM registers
7445 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7446 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7449 //===----------------------------------------------------------------------===//
7450 // Half precision conversion instructions
7451 //===----------------------------------------------------------------------===//
7452 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7453 let Predicates = [HasAVX, HasF16C] in {
7454 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7455 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7456 [(set RC:$dst, (Int VR128:$src))]>,
7458 let neverHasSideEffects = 1, mayLoad = 1 in
7459 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7460 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7464 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7465 let Predicates = [HasAVX, HasF16C] in {
7466 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7467 (ins RC:$src1, i32i8imm:$src2),
7468 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7469 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7471 let neverHasSideEffects = 1, mayLoad = 1 in
7472 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7473 (ins RC:$src1, i32i8imm:$src2),
7474 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7479 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7480 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7481 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7482 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7484 //===----------------------------------------------------------------------===//
7485 // AVX2 Instructions
7486 //===----------------------------------------------------------------------===//
7488 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7489 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7490 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7491 X86MemOperand x86memop> {
7492 let isCommutable = 1 in
7493 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7494 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7495 !strconcat(OpcodeStr,
7496 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7497 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7499 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7500 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7501 !strconcat(OpcodeStr,
7502 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7505 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7509 let isCommutable = 0 in {
7510 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7511 VR128, memopv2i64, i128mem>;
7512 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7513 VR256, memopv4i64, i256mem>;
7516 //===----------------------------------------------------------------------===//
7517 // VPBROADCAST - Load from memory and broadcast to all elements of the
7518 // destination operand
7520 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7521 X86MemOperand x86memop, PatFrag ld_frag,
7522 Intrinsic Int128, Intrinsic Int256> {
7523 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7524 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7525 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7526 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7527 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7529 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7530 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7531 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7532 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7533 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7534 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7536 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7539 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7540 int_x86_avx2_pbroadcastb_128,
7541 int_x86_avx2_pbroadcastb_256>;
7542 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7543 int_x86_avx2_pbroadcastw_128,
7544 int_x86_avx2_pbroadcastw_256>;
7545 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7546 int_x86_avx2_pbroadcastd_128,
7547 int_x86_avx2_pbroadcastd_256>;
7548 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7549 int_x86_avx2_pbroadcastq_128,
7550 int_x86_avx2_pbroadcastq_256>;
7552 let Predicates = [HasAVX2] in {
7553 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7554 (VPBROADCASTBrm addr:$src)>;
7555 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7556 (VPBROADCASTBYrm addr:$src)>;
7557 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7558 (VPBROADCASTWrm addr:$src)>;
7559 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7560 (VPBROADCASTWYrm addr:$src)>;
7561 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7562 (VPBROADCASTDrm addr:$src)>;
7563 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7564 (VPBROADCASTDYrm addr:$src)>;
7565 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7566 (VPBROADCASTQrm addr:$src)>;
7567 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7568 (VPBROADCASTQYrm addr:$src)>;
7571 // AVX1 broadcast patterns
7572 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7573 (VBROADCASTSSYrm addr:$src)>;
7574 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7575 (VBROADCASTSDrm addr:$src)>;
7576 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7577 (VBROADCASTSSYrm addr:$src)>;
7578 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7579 (VBROADCASTSDrm addr:$src)>;
7581 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7582 (VBROADCASTSSrm addr:$src)>;
7583 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7584 (VBROADCASTSSrm addr:$src)>;
7586 //===----------------------------------------------------------------------===//
7587 // VPERM - Permute instructions
7590 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7592 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7593 (ins VR256:$src1, VR256:$src2),
7594 !strconcat(OpcodeStr,
7595 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7596 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7597 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7598 (ins VR256:$src1, i256mem:$src2),
7599 !strconcat(OpcodeStr,
7600 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7601 [(set VR256:$dst, (Int VR256:$src1,
7602 (bitconvert (mem_frag addr:$src2))))]>,
7606 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7607 let ExeDomain = SSEPackedSingle in
7608 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7610 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7612 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7613 (ins VR256:$src1, i8imm:$src2),
7614 !strconcat(OpcodeStr,
7615 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7616 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7617 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7618 (ins i256mem:$src1, i8imm:$src2),
7619 !strconcat(OpcodeStr,
7620 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7621 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7625 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7627 let ExeDomain = SSEPackedDouble in
7628 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7631 //===----------------------------------------------------------------------===//
7632 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7634 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7635 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7636 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7638 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7640 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7641 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7642 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7644 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7648 let Predicates = [HasAVX2] in {
7649 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7650 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7651 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7652 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7653 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7654 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7655 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7656 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7658 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7660 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7661 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7662 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7663 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7664 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7666 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7667 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7669 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7673 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7674 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7675 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7676 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7677 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7678 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7679 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7680 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7681 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7682 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7683 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7684 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7686 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7687 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7688 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7689 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7690 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7691 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7692 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7693 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7694 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7695 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7696 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7697 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7698 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7699 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7700 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7701 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7702 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7703 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7706 //===----------------------------------------------------------------------===//
7707 // VINSERTI128 - Insert packed integer values
7709 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7710 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7711 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7713 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7715 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7716 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7717 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7719 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7720 imm:$src3))]>, VEX_4V;
7722 let Predicates = [HasAVX2] in {
7723 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7725 (VINSERTI128rr VR256:$src1, VR128:$src2,
7726 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7727 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7729 (VINSERTI128rr VR256:$src1, VR128:$src2,
7730 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7731 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7733 (VINSERTI128rr VR256:$src1, VR128:$src2,
7734 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7735 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7737 (VINSERTI128rr VR256:$src1, VR128:$src2,
7738 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7742 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7744 (VINSERTF128rr VR256:$src1, VR128:$src2,
7745 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7746 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7748 (VINSERTF128rr VR256:$src1, VR128:$src2,
7749 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7750 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7752 (VINSERTF128rr VR256:$src1, VR128:$src2,
7753 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7754 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7756 (VINSERTF128rr VR256:$src1, VR128:$src2,
7757 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7758 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7760 (VINSERTF128rr VR256:$src1, VR128:$src2,
7761 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7762 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7764 (VINSERTF128rr VR256:$src1, VR128:$src2,
7765 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7767 //===----------------------------------------------------------------------===//
7768 // VEXTRACTI128 - Extract packed integer values
7770 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7771 (ins VR256:$src1, i8imm:$src2),
7772 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7774 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7776 let neverHasSideEffects = 1, mayStore = 1 in
7777 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7778 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7779 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7781 let Predicates = [HasAVX2] in {
7782 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7783 (v2i64 (VEXTRACTI128rr
7784 (v4i64 VR256:$src1),
7785 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7786 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7787 (v4i32 (VEXTRACTI128rr
7788 (v8i32 VR256:$src1),
7789 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7790 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7791 (v8i16 (VEXTRACTI128rr
7792 (v16i16 VR256:$src1),
7793 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7794 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7795 (v16i8 (VEXTRACTI128rr
7796 (v32i8 VR256:$src1),
7797 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7801 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7802 (v4f32 (VEXTRACTF128rr
7803 (v8f32 VR256:$src1),
7804 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7805 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7806 (v2f64 (VEXTRACTF128rr
7807 (v4f64 VR256:$src1),
7808 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7809 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7810 (v2i64 (VEXTRACTF128rr
7811 (v4i64 VR256:$src1),
7812 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7813 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7814 (v4i32 (VEXTRACTF128rr
7815 (v8i32 VR256:$src1),
7816 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7817 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7818 (v8i16 (VEXTRACTF128rr
7819 (v16i16 VR256:$src1),
7820 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7821 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7822 (v16i8 (VEXTRACTF128rr
7823 (v32i8 VR256:$src1),
7824 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7826 //===----------------------------------------------------------------------===//
7827 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7829 multiclass avx2_pmovmask<string OpcodeStr,
7830 Intrinsic IntLd128, Intrinsic IntLd256,
7831 Intrinsic IntSt128, Intrinsic IntSt256> {
7832 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7833 (ins VR128:$src1, i128mem:$src2),
7834 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7835 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7836 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7837 (ins VR256:$src1, i256mem:$src2),
7838 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7839 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7840 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7841 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7842 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7843 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7844 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7845 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7846 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7847 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7850 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7851 int_x86_avx2_maskload_d,
7852 int_x86_avx2_maskload_d_256,
7853 int_x86_avx2_maskstore_d,
7854 int_x86_avx2_maskstore_d_256>;
7855 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7856 int_x86_avx2_maskload_q,
7857 int_x86_avx2_maskload_q_256,
7858 int_x86_avx2_maskstore_q,
7859 int_x86_avx2_maskstore_q_256>, VEX_W;
7862 //===----------------------------------------------------------------------===//
7863 // Variable Bit Shifts
7865 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7866 ValueType vt128, ValueType vt256> {
7867 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7868 (ins VR128:$src1, VR128:$src2),
7869 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7871 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7873 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7874 (ins VR128:$src1, i128mem:$src2),
7875 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7877 (vt128 (OpNode VR128:$src1,
7878 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7880 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7881 (ins VR256:$src1, VR256:$src2),
7882 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7884 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7886 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7887 (ins VR256:$src1, i256mem:$src2),
7888 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7890 (vt256 (OpNode VR256:$src1,
7891 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7895 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7896 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7897 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7898 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7899 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;