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:$src)>;
566 let AddedComplexity = 20 in {
567 // MOVSSrm zeros the high parts of the register; represent this
568 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
569 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
570 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
571 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
572 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
573 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
574 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
576 // MOVSDrm zeros the high parts of the register; represent this
577 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
578 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
579 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
580 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
581 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
582 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
583 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
584 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
585 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
586 def : Pat<(v2f64 (X86vzload addr:$src)),
587 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
589 // Represent the same patterns above but in the form they appear for
591 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
592 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
593 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
594 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
595 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
596 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
598 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
599 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
600 (SUBREG_TO_REG (i32 0),
601 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
603 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
604 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
605 (SUBREG_TO_REG (i64 0),
606 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
609 // Extract and store.
610 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
613 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
614 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
617 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
619 // Shuffle with VMOVSS
620 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
621 (VMOVSSrr VR128:$src1, FR32:$src2)>;
622 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
623 (VMOVSSrr (v4i32 VR128:$src1),
624 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
625 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
626 (VMOVSSrr (v4f32 VR128:$src1),
627 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
629 // Shuffle with VMOVSD
630 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
631 (VMOVSDrr VR128:$src1, FR64:$src2)>;
632 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
633 (VMOVSDrr (v2i64 VR128:$src1),
634 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
635 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
636 (VMOVSDrr (v2f64 VR128:$src1),
637 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
638 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
639 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
641 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
642 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
645 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
646 // is during lowering, where it's not possible to recognize the fold cause
647 // it has two uses through a bitcast. One use disappears at isel time and the
648 // fold opportunity reappears.
649 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
650 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
652 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
653 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
655 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
656 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
658 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
659 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
663 //===----------------------------------------------------------------------===//
664 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
665 //===----------------------------------------------------------------------===//
667 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
668 X86MemOperand x86memop, PatFrag ld_frag,
669 string asm, Domain d,
670 bit IsReMaterializable = 1> {
671 let neverHasSideEffects = 1 in
672 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
673 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
674 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
675 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
676 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
677 [(set RC:$dst, (ld_frag addr:$src))], d>;
680 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
681 "movaps", SSEPackedSingle>, TB, VEX;
682 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
683 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
684 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
685 "movups", SSEPackedSingle>, TB, VEX;
686 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
687 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
689 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
690 "movaps", SSEPackedSingle>, TB, VEX;
691 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
692 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
693 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
694 "movups", SSEPackedSingle>, TB, VEX;
695 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
696 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
697 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
698 "movaps", SSEPackedSingle>, TB;
699 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
700 "movapd", SSEPackedDouble>, TB, OpSize;
701 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
702 "movups", SSEPackedSingle>, TB;
703 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
704 "movupd", SSEPackedDouble, 0>, TB, OpSize;
706 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
707 "movaps\t{$src, $dst|$dst, $src}",
708 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
709 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
710 "movapd\t{$src, $dst|$dst, $src}",
711 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
712 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
713 "movups\t{$src, $dst|$dst, $src}",
714 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
715 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
716 "movupd\t{$src, $dst|$dst, $src}",
717 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
718 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
719 "movaps\t{$src, $dst|$dst, $src}",
720 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
721 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
722 "movapd\t{$src, $dst|$dst, $src}",
723 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
724 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
725 "movups\t{$src, $dst|$dst, $src}",
726 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
727 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
728 "movupd\t{$src, $dst|$dst, $src}",
729 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
732 let isCodeGenOnly = 1 in {
733 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
735 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
736 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
738 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
739 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
741 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
742 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
744 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
745 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
747 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
748 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
750 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
751 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
753 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
754 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
756 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
759 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
760 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
761 (VMOVUPSYmr addr:$dst, VR256:$src)>;
763 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
764 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
765 (VMOVUPDYmr addr:$dst, VR256:$src)>;
767 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
768 "movaps\t{$src, $dst|$dst, $src}",
769 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
770 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
771 "movapd\t{$src, $dst|$dst, $src}",
772 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
773 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
774 "movups\t{$src, $dst|$dst, $src}",
775 [(store (v4f32 VR128:$src), addr:$dst)]>;
776 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
777 "movupd\t{$src, $dst|$dst, $src}",
778 [(store (v2f64 VR128:$src), addr:$dst)]>;
781 let isCodeGenOnly = 1 in {
782 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
783 "movaps\t{$src, $dst|$dst, $src}", []>;
784 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
785 "movapd\t{$src, $dst|$dst, $src}", []>;
786 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
787 "movups\t{$src, $dst|$dst, $src}", []>;
788 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
789 "movupd\t{$src, $dst|$dst, $src}", []>;
792 let Predicates = [HasAVX] in {
793 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
794 (VMOVUPSmr addr:$dst, VR128:$src)>;
795 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
796 (VMOVUPDmr addr:$dst, VR128:$src)>;
799 let Predicates = [HasSSE1] in
800 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
801 (MOVUPSmr addr:$dst, VR128:$src)>;
802 let Predicates = [HasSSE2] in
803 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
804 (MOVUPDmr addr:$dst, VR128:$src)>;
806 // Use movaps / movups for SSE integer load / store (one byte shorter).
807 // The instructions selected below are then converted to MOVDQA/MOVDQU
808 // during the SSE domain pass.
809 let Predicates = [HasSSE1] in {
810 def : Pat<(alignedloadv4i32 addr:$src),
811 (MOVAPSrm addr:$src)>;
812 def : Pat<(loadv4i32 addr:$src),
813 (MOVUPSrm addr:$src)>;
814 def : Pat<(alignedloadv2i64 addr:$src),
815 (MOVAPSrm addr:$src)>;
816 def : Pat<(loadv2i64 addr:$src),
817 (MOVUPSrm addr:$src)>;
819 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
820 (MOVAPSmr addr:$dst, VR128:$src)>;
821 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
822 (MOVAPSmr addr:$dst, VR128:$src)>;
823 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
824 (MOVAPSmr addr:$dst, VR128:$src)>;
825 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
826 (MOVAPSmr addr:$dst, VR128:$src)>;
827 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
828 (MOVUPSmr addr:$dst, VR128:$src)>;
829 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
830 (MOVUPSmr addr:$dst, VR128:$src)>;
831 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
832 (MOVUPSmr addr:$dst, VR128:$src)>;
833 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
834 (MOVUPSmr addr:$dst, VR128:$src)>;
837 // Use vmovaps/vmovups for AVX integer load/store.
838 let Predicates = [HasAVX] in {
839 // 128-bit load/store
840 def : Pat<(alignedloadv4i32 addr:$src),
841 (VMOVAPSrm addr:$src)>;
842 def : Pat<(loadv4i32 addr:$src),
843 (VMOVUPSrm addr:$src)>;
844 def : Pat<(alignedloadv2i64 addr:$src),
845 (VMOVAPSrm addr:$src)>;
846 def : Pat<(loadv2i64 addr:$src),
847 (VMOVUPSrm addr:$src)>;
849 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
850 (VMOVAPSmr addr:$dst, VR128:$src)>;
851 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
852 (VMOVAPSmr addr:$dst, VR128:$src)>;
853 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
854 (VMOVAPSmr addr:$dst, VR128:$src)>;
855 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
856 (VMOVAPSmr addr:$dst, VR128:$src)>;
857 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
858 (VMOVUPSmr addr:$dst, VR128:$src)>;
859 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
860 (VMOVUPSmr addr:$dst, VR128:$src)>;
861 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
862 (VMOVUPSmr addr:$dst, VR128:$src)>;
863 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
864 (VMOVUPSmr addr:$dst, VR128:$src)>;
866 // 256-bit load/store
867 def : Pat<(alignedloadv4i64 addr:$src),
868 (VMOVAPSYrm addr:$src)>;
869 def : Pat<(loadv4i64 addr:$src),
870 (VMOVUPSYrm addr:$src)>;
871 def : Pat<(alignedloadv8i32 addr:$src),
872 (VMOVAPSYrm addr:$src)>;
873 def : Pat<(loadv8i32 addr:$src),
874 (VMOVUPSYrm addr:$src)>;
875 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
876 (VMOVAPSYmr addr:$dst, VR256:$src)>;
877 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
878 (VMOVAPSYmr addr:$dst, VR256:$src)>;
879 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
880 (VMOVAPSYmr addr:$dst, VR256:$src)>;
881 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
882 (VMOVAPSYmr addr:$dst, VR256:$src)>;
883 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
884 (VMOVUPSYmr addr:$dst, VR256:$src)>;
885 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
886 (VMOVUPSYmr addr:$dst, VR256:$src)>;
887 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
888 (VMOVUPSYmr addr:$dst, VR256:$src)>;
889 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
890 (VMOVUPSYmr addr:$dst, VR256:$src)>;
893 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
894 // bits are disregarded. FIXME: Set encoding to pseudo!
895 let neverHasSideEffects = 1 in {
896 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
897 "movaps\t{$src, $dst|$dst, $src}", []>;
898 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
899 "movapd\t{$src, $dst|$dst, $src}", []>;
900 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
901 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
902 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
903 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
906 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
907 // bits are disregarded. FIXME: Set encoding to pseudo!
908 let canFoldAsLoad = 1, isReMaterializable = 1 in {
909 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
910 "movaps\t{$src, $dst|$dst, $src}",
911 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
912 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
913 "movapd\t{$src, $dst|$dst, $src}",
914 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
915 let isCodeGenOnly = 1 in {
916 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
917 "movaps\t{$src, $dst|$dst, $src}",
918 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
919 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
920 "movapd\t{$src, $dst|$dst, $src}",
921 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
925 //===----------------------------------------------------------------------===//
926 // SSE 1 & 2 - Move Low packed FP Instructions
927 //===----------------------------------------------------------------------===//
929 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
930 PatFrag mov_frag, string base_opc,
932 def PSrm : PI<opc, MRMSrcMem,
933 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
934 !strconcat(base_opc, "s", asm_opr),
937 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
938 SSEPackedSingle>, TB;
940 def PDrm : PI<opc, MRMSrcMem,
941 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
942 !strconcat(base_opc, "d", asm_opr),
943 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
944 (scalar_to_vector (loadf64 addr:$src2)))))],
945 SSEPackedDouble>, TB, OpSize;
948 let AddedComplexity = 20 in {
949 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
950 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
952 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
953 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
954 "\t{$src2, $dst|$dst, $src2}">;
957 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
958 "movlps\t{$src, $dst|$dst, $src}",
959 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
960 (iPTR 0))), addr:$dst)]>, VEX;
961 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
962 "movlpd\t{$src, $dst|$dst, $src}",
963 [(store (f64 (vector_extract (v2f64 VR128:$src),
964 (iPTR 0))), addr:$dst)]>, VEX;
965 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
966 "movlps\t{$src, $dst|$dst, $src}",
967 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
968 (iPTR 0))), addr:$dst)]>;
969 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
970 "movlpd\t{$src, $dst|$dst, $src}",
971 [(store (f64 (vector_extract (v2f64 VR128:$src),
972 (iPTR 0))), addr:$dst)]>;
974 let Predicates = [HasAVX] in {
975 let AddedComplexity = 20 in {
976 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
977 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
978 (VMOVLPSrm VR128:$src1, addr:$src2)>;
979 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
980 (VMOVLPSrm VR128:$src1, addr:$src2)>;
981 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
982 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
983 (VMOVLPDrm VR128:$src1, addr:$src2)>;
984 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
985 (VMOVLPDrm VR128:$src1, addr:$src2)>;
988 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
989 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
990 (VMOVLPSmr addr:$src1, VR128:$src2)>;
991 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
992 VR128:$src2)), addr:$src1),
993 (VMOVLPSmr addr:$src1, VR128:$src2)>;
995 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
996 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
997 (VMOVLPDmr addr:$src1, VR128:$src2)>;
998 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
999 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1001 // Shuffle with VMOVLPS
1002 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1003 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1004 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1005 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1006 def : Pat<(X86Movlps VR128:$src1,
1007 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1008 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1010 // Shuffle with VMOVLPD
1011 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1012 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1013 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1014 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1015 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1016 (scalar_to_vector (loadf64 addr:$src2)))),
1017 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1020 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1022 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1023 def : Pat<(store (v4i32 (X86Movlps
1024 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1025 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1026 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1028 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1029 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1031 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1034 let Predicates = [HasSSE1] in {
1035 let AddedComplexity = 20 in {
1036 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1037 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1038 (MOVLPSrm VR128:$src1, addr:$src2)>;
1039 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1040 (MOVLPSrm VR128:$src1, addr:$src2)>;
1043 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1044 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1045 (iPTR 0))), addr:$src1),
1046 (MOVLPSmr addr:$src1, VR128:$src2)>;
1047 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1048 (MOVLPSmr addr:$src1, VR128:$src2)>;
1049 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1050 VR128:$src2)), addr:$src1),
1051 (MOVLPSmr addr:$src1, VR128:$src2)>;
1053 // Shuffle with MOVLPS
1054 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1055 (MOVLPSrm VR128:$src1, addr:$src2)>;
1056 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1057 (MOVLPSrm VR128:$src1, addr:$src2)>;
1058 def : Pat<(X86Movlps VR128:$src1,
1059 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1060 (MOVLPSrm VR128:$src1, addr:$src2)>;
1061 def : Pat<(X86Movlps VR128:$src1,
1062 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1063 (MOVLPSrm VR128:$src1, addr:$src2)>;
1066 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1068 (MOVLPSmr addr:$src1, VR128:$src2)>;
1069 def : Pat<(store (v4i32 (X86Movlps
1070 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1072 (MOVLPSmr addr:$src1, VR128:$src2)>;
1075 let Predicates = [HasSSE2] in {
1076 let AddedComplexity = 20 in {
1077 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1078 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1079 (MOVLPDrm VR128:$src1, addr:$src2)>;
1080 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1081 (MOVLPDrm VR128:$src1, addr:$src2)>;
1084 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1085 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1086 (MOVLPDmr addr:$src1, VR128:$src2)>;
1087 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1088 (MOVLPDmr addr:$src1, VR128:$src2)>;
1090 // Shuffle with MOVLPD
1091 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1092 (MOVLPDrm VR128:$src1, addr:$src2)>;
1093 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1094 (MOVLPDrm VR128:$src1, addr:$src2)>;
1095 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1096 (scalar_to_vector (loadf64 addr:$src2)))),
1097 (MOVLPDrm VR128:$src1, addr:$src2)>;
1100 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1102 (MOVLPDmr addr:$src1, VR128:$src2)>;
1103 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1105 (MOVLPDmr addr:$src1, VR128:$src2)>;
1108 //===----------------------------------------------------------------------===//
1109 // SSE 1 & 2 - Move Hi packed FP Instructions
1110 //===----------------------------------------------------------------------===//
1112 let AddedComplexity = 20 in {
1113 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1114 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1116 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1117 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1118 "\t{$src2, $dst|$dst, $src2}">;
1121 // v2f64 extract element 1 is always custom lowered to unpack high to low
1122 // and extract element 0 so the non-store version isn't too horrible.
1123 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1124 "movhps\t{$src, $dst|$dst, $src}",
1125 [(store (f64 (vector_extract
1126 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1127 (undef)), (iPTR 0))), addr:$dst)]>,
1129 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1130 "movhpd\t{$src, $dst|$dst, $src}",
1131 [(store (f64 (vector_extract
1132 (v2f64 (unpckh VR128:$src, (undef))),
1133 (iPTR 0))), addr:$dst)]>,
1135 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1136 "movhps\t{$src, $dst|$dst, $src}",
1137 [(store (f64 (vector_extract
1138 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1139 (undef)), (iPTR 0))), addr:$dst)]>;
1140 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1141 "movhpd\t{$src, $dst|$dst, $src}",
1142 [(store (f64 (vector_extract
1143 (v2f64 (unpckh VR128:$src, (undef))),
1144 (iPTR 0))), addr:$dst)]>;
1146 let Predicates = [HasAVX] in {
1148 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1149 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1150 def : Pat<(X86Movlhps VR128:$src1,
1151 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1152 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1153 def : Pat<(X86Movlhps VR128:$src1,
1154 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1155 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1156 def : Pat<(X86Movlhps VR128:$src1,
1157 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1158 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1160 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1161 // is during lowering, where it's not possible to recognize the load fold cause
1162 // it has two uses through a bitcast. One use disappears at isel time and the
1163 // fold opportunity reappears.
1164 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1165 (scalar_to_vector (loadf64 addr:$src2)))),
1166 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1168 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1169 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1170 (scalar_to_vector (loadf64 addr:$src2)))),
1171 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1174 def : Pat<(store (f64 (vector_extract
1175 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1176 (VMOVHPSmr addr:$dst, VR128:$src)>;
1177 def : Pat<(store (f64 (vector_extract
1178 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1179 (VMOVHPDmr addr:$dst, VR128:$src)>;
1182 let Predicates = [HasSSE1] in {
1184 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1185 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1186 def : Pat<(X86Movlhps VR128:$src1,
1187 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1188 (MOVHPSrm VR128:$src1, addr:$src2)>;
1189 def : Pat<(X86Movlhps VR128:$src1,
1190 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1191 (MOVHPSrm VR128:$src1, addr:$src2)>;
1192 def : Pat<(X86Movlhps VR128:$src1,
1193 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1194 (MOVHPSrm VR128:$src1, addr:$src2)>;
1197 def : Pat<(store (f64 (vector_extract
1198 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1199 (MOVHPSmr addr:$dst, VR128:$src)>;
1202 let Predicates = [HasSSE2] in {
1203 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1204 // is during lowering, where it's not possible to recognize the load fold cause
1205 // it has two uses through a bitcast. One use disappears at isel time and the
1206 // fold opportunity reappears.
1207 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1208 (scalar_to_vector (loadf64 addr:$src2)))),
1209 (MOVHPDrm VR128:$src1, addr:$src2)>;
1211 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1212 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1213 (scalar_to_vector (loadf64 addr:$src2)))),
1214 (MOVHPDrm VR128:$src1, addr:$src2)>;
1217 def : Pat<(store (f64 (vector_extract
1218 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1219 (MOVHPDmr addr:$dst, VR128:$src)>;
1222 //===----------------------------------------------------------------------===//
1223 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1224 //===----------------------------------------------------------------------===//
1226 let AddedComplexity = 20 in {
1227 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1228 (ins VR128:$src1, VR128:$src2),
1229 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1231 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1233 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1234 (ins VR128:$src1, VR128:$src2),
1235 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1237 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1240 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1241 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1242 (ins VR128:$src1, VR128:$src2),
1243 "movlhps\t{$src2, $dst|$dst, $src2}",
1245 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1246 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1247 (ins VR128:$src1, VR128:$src2),
1248 "movhlps\t{$src2, $dst|$dst, $src2}",
1250 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1253 let Predicates = [HasAVX] in {
1255 let AddedComplexity = 20 in {
1256 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1257 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1258 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1259 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1261 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1262 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1263 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1265 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1266 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1267 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1268 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1269 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1270 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1273 let AddedComplexity = 20 in {
1274 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1275 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1276 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1278 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1279 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1280 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1281 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1282 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1285 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1286 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1287 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1288 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1291 let Predicates = [HasSSE1] in {
1293 let AddedComplexity = 20 in {
1294 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1295 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1296 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1297 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1299 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1300 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1301 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1303 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1304 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1305 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1306 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1307 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1308 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1311 let AddedComplexity = 20 in {
1312 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1313 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1314 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1316 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1317 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1318 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1319 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1320 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1323 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1324 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1325 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1326 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1329 //===----------------------------------------------------------------------===//
1330 // SSE 1 & 2 - Conversion Instructions
1331 //===----------------------------------------------------------------------===//
1333 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1334 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1336 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1337 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1338 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1339 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1342 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1343 X86MemOperand x86memop, string asm> {
1344 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1346 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1349 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1350 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1351 string asm, Domain d> {
1352 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1353 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1354 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1355 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1358 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1359 X86MemOperand x86memop, string asm> {
1360 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1361 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1363 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1364 (ins DstRC:$src1, x86memop:$src),
1365 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1368 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1369 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1371 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1372 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1374 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1375 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1377 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1378 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1379 VEX, VEX_W, VEX_LIG;
1381 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1382 // register, but the same isn't true when only using memory operands,
1383 // provide other assembly "l" and "q" forms to address this explicitly
1384 // where appropriate to do so.
1385 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1387 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1388 VEX_4V, VEX_W, VEX_LIG;
1389 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1391 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1393 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1394 VEX_4V, VEX_W, VEX_LIG;
1396 let Predicates = [HasAVX] in {
1397 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1398 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1399 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1400 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1401 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1402 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1403 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1404 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1406 def : Pat<(f32 (sint_to_fp GR32:$src)),
1407 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1408 def : Pat<(f32 (sint_to_fp GR64:$src)),
1409 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1410 def : Pat<(f64 (sint_to_fp GR32:$src)),
1411 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1412 def : Pat<(f64 (sint_to_fp GR64:$src)),
1413 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1416 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1417 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1418 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1419 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1420 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1421 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1422 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1423 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1424 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1425 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1426 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1427 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1428 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1429 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1430 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1431 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1433 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1434 // and/or XMM operand(s).
1436 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1437 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1439 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1440 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1441 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1442 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1443 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1444 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1447 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1448 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1449 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1450 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1452 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1453 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1454 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1455 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1456 (ins DstRC:$src1, x86memop:$src2),
1458 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1459 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1460 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1463 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1464 f128mem, load, "cvtsd2si">, XD, VEX;
1465 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1466 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1469 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1470 // Get rid of this hack or rename the intrinsics, there are several
1471 // intructions that only match with the intrinsic form, why create duplicates
1472 // to let them be recognized by the assembler?
1473 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1474 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1475 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1476 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1479 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1480 f128mem, load, "cvtsd2si{l}">, XD;
1481 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1482 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1485 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1486 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1487 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1488 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1490 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1491 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1492 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1493 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1496 let Constraints = "$src1 = $dst" in {
1497 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1498 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1500 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1501 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1502 "cvtsi2ss{q}">, XS, REX_W;
1503 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1504 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1506 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1507 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1508 "cvtsi2sd">, XD, REX_W;
1513 // Aliases for intrinsics
1514 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1515 f32mem, load, "cvttss2si">, XS, VEX;
1516 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1517 int_x86_sse_cvttss2si64, f32mem, load,
1518 "cvttss2si">, XS, VEX, VEX_W;
1519 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1520 f128mem, load, "cvttsd2si">, XD, VEX;
1521 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1522 int_x86_sse2_cvttsd2si64, f128mem, load,
1523 "cvttsd2si">, XD, VEX, VEX_W;
1524 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1525 f32mem, load, "cvttss2si">, XS;
1526 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1527 int_x86_sse_cvttss2si64, f32mem, load,
1528 "cvttss2si{q}">, XS, REX_W;
1529 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1530 f128mem, load, "cvttsd2si">, XD;
1531 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1532 int_x86_sse2_cvttsd2si64, f128mem, load,
1533 "cvttsd2si{q}">, XD, REX_W;
1535 let Pattern = []<dag> in {
1536 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1537 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1539 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1540 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1542 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1543 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1544 SSEPackedSingle>, TB, VEX;
1545 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1546 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1547 SSEPackedSingle>, TB, VEX;
1550 let Pattern = []<dag> in {
1551 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1552 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1553 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1554 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1555 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1556 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1557 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1560 let Predicates = [HasSSE1] in {
1561 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1562 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1563 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1564 (CVTSS2SIrm addr:$src)>;
1565 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1566 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1567 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1568 (CVTSS2SI64rm addr:$src)>;
1571 let Predicates = [HasAVX] in {
1572 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1573 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1574 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1575 (VCVTSS2SIrm addr:$src)>;
1576 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1577 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1578 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1579 (VCVTSS2SI64rm addr:$src)>;
1584 // Convert scalar double to scalar single
1585 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1586 (ins FR64:$src1, FR64:$src2),
1587 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1590 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1591 (ins FR64:$src1, f64mem:$src2),
1592 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1593 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1595 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1598 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1599 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1600 [(set FR32:$dst, (fround FR64:$src))]>;
1601 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1602 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1603 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1604 Requires<[HasSSE2, OptForSize]>;
1606 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1607 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1609 let Constraints = "$src1 = $dst" in
1610 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1611 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1613 // Convert scalar single to scalar double
1614 // SSE2 instructions with XS prefix
1615 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1616 (ins FR32:$src1, FR32:$src2),
1617 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1618 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1620 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1621 (ins FR32:$src1, f32mem:$src2),
1622 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1623 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1625 let Predicates = [HasAVX] in {
1626 def : Pat<(f64 (fextend FR32:$src)),
1627 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1628 def : Pat<(fextend (loadf32 addr:$src)),
1629 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1630 def : Pat<(extloadf32 addr:$src),
1631 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1634 def : Pat<(extloadf32 addr:$src),
1635 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1636 Requires<[HasAVX, OptForSpeed]>;
1638 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1639 "cvtss2sd\t{$src, $dst|$dst, $src}",
1640 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1641 Requires<[HasSSE2]>;
1642 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1643 "cvtss2sd\t{$src, $dst|$dst, $src}",
1644 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1645 Requires<[HasSSE2, OptForSize]>;
1647 // extload f32 -> f64. This matches load+fextend because we have a hack in
1648 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1650 // Since these loads aren't folded into the fextend, we have to match it
1652 def : Pat<(fextend (loadf32 addr:$src)),
1653 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1654 def : Pat<(extloadf32 addr:$src),
1655 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1657 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1658 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1659 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1660 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1661 VR128:$src2))]>, XS, VEX_4V,
1663 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1664 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1665 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1666 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1667 (load addr:$src2)))]>, XS, VEX_4V,
1669 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1670 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1671 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1672 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1673 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1674 VR128:$src2))]>, XS,
1675 Requires<[HasSSE2]>;
1676 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1677 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1678 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1679 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1680 (load addr:$src2)))]>, XS,
1681 Requires<[HasSSE2]>;
1684 // Convert doubleword to packed single/double fp
1685 // SSE2 instructions without OpSize prefix
1686 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1687 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1688 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1689 TB, VEX, Requires<[HasAVX]>;
1690 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1691 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1692 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1693 (bitconvert (memopv2i64 addr:$src))))]>,
1694 TB, VEX, Requires<[HasAVX]>;
1695 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1696 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1697 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1698 TB, Requires<[HasSSE2]>;
1699 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1700 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1701 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1702 (bitconvert (memopv2i64 addr:$src))))]>,
1703 TB, Requires<[HasSSE2]>;
1705 // FIXME: why the non-intrinsic version is described as SSE3?
1706 // SSE2 instructions with XS prefix
1707 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1708 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1709 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1710 XS, VEX, Requires<[HasAVX]>;
1711 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1712 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1713 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1714 (bitconvert (memopv2i64 addr:$src))))]>,
1715 XS, VEX, Requires<[HasAVX]>;
1716 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1717 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1718 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1719 XS, Requires<[HasSSE2]>;
1720 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1721 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1722 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1723 (bitconvert (memopv2i64 addr:$src))))]>,
1724 XS, Requires<[HasSSE2]>;
1727 // Convert packed single/double fp to doubleword
1728 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1729 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1730 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1731 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1732 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1733 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1734 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1735 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1736 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1737 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1738 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1739 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1741 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1742 "cvtps2dq\t{$src, $dst|$dst, $src}",
1743 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1745 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1747 "cvtps2dq\t{$src, $dst|$dst, $src}",
1748 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1749 (memop addr:$src)))]>, VEX;
1750 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1751 "cvtps2dq\t{$src, $dst|$dst, $src}",
1752 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1753 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1754 "cvtps2dq\t{$src, $dst|$dst, $src}",
1755 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1756 (memop addr:$src)))]>;
1758 // SSE2 packed instructions with XD prefix
1759 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1760 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1762 XD, VEX, Requires<[HasAVX]>;
1763 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1764 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1766 (memop addr:$src)))]>,
1767 XD, VEX, Requires<[HasAVX]>;
1768 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1769 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1770 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1771 XD, Requires<[HasSSE2]>;
1772 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1773 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1774 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1775 (memop addr:$src)))]>,
1776 XD, Requires<[HasSSE2]>;
1779 // Convert with truncation packed single/double fp to doubleword
1780 // SSE2 packed instructions with XS prefix
1781 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1782 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1784 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1785 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1786 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1787 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1789 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1790 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1791 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1792 "cvttps2dq\t{$src, $dst|$dst, $src}",
1794 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1795 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1796 "cvttps2dq\t{$src, $dst|$dst, $src}",
1798 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1800 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1801 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1803 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1804 XS, VEX, Requires<[HasAVX]>;
1805 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1806 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1807 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1808 (memop addr:$src)))]>,
1809 XS, VEX, Requires<[HasAVX]>;
1811 let Predicates = [HasSSE2] in {
1812 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1813 (Int_CVTDQ2PSrr VR128:$src)>;
1814 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1815 (CVTTPS2DQrr VR128:$src)>;
1818 let Predicates = [HasAVX] in {
1819 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1820 (Int_VCVTDQ2PSrr VR128:$src)>;
1821 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1822 (VCVTTPS2DQrr VR128:$src)>;
1823 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1824 (VCVTDQ2PSYrr VR256:$src)>;
1825 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1826 (VCVTTPS2DQYrr VR256:$src)>;
1829 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1830 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1832 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1833 let isCodeGenOnly = 1 in
1834 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1835 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1836 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1837 (memop addr:$src)))]>, VEX;
1838 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1839 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1840 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1841 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1842 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1843 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1844 (memop addr:$src)))]>;
1846 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1847 // register, but the same isn't true when using memory operands instead.
1848 // Provide other assembly rr and rm forms to address this explicitly.
1849 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1850 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1854 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1855 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1856 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1859 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1860 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1861 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1862 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1864 // Convert packed single to packed double
1865 let Predicates = [HasAVX] in {
1866 // SSE2 instructions without OpSize prefix
1867 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1868 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1869 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1870 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1871 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1872 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1873 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1874 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1876 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1877 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1878 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1879 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1881 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1882 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1883 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1884 TB, VEX, Requires<[HasAVX]>;
1885 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1886 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1887 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1888 (load addr:$src)))]>,
1889 TB, VEX, Requires<[HasAVX]>;
1890 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1891 "cvtps2pd\t{$src, $dst|$dst, $src}",
1892 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1893 TB, Requires<[HasSSE2]>;
1894 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1895 "cvtps2pd\t{$src, $dst|$dst, $src}",
1896 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1897 (load addr:$src)))]>,
1898 TB, Requires<[HasSSE2]>;
1900 // Convert packed double to packed single
1901 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1902 // register, but the same isn't true when using memory operands instead.
1903 // Provide other assembly rr and rm forms to address this explicitly.
1904 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1905 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1906 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1907 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1910 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1911 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1912 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1913 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1916 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1917 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1918 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1919 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1920 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1921 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1922 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1923 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1926 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1927 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1928 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1929 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1931 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1932 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1933 (memop addr:$src)))]>;
1934 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1935 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1936 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1937 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1938 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1939 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1940 (memop addr:$src)))]>;
1942 // AVX 256-bit register conversion intrinsics
1943 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
1944 // whenever possible to avoid declaring two versions of each one.
1945 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
1946 (VCVTDQ2PSYrr VR256:$src)>;
1947 def : Pat<(int_x86_avx_cvtdq2_ps_256 (memopv8i32 addr:$src)),
1948 (VCVTDQ2PSYrm addr:$src)>;
1950 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
1951 (VCVTPD2PSYrr VR256:$src)>;
1952 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
1953 (VCVTPD2PSYrm addr:$src)>;
1955 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
1956 (VCVTPS2DQYrr VR256:$src)>;
1957 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
1958 (VCVTPS2DQYrm addr:$src)>;
1960 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
1961 (VCVTPS2PDYrr VR128:$src)>;
1962 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
1963 (VCVTPS2PDYrm addr:$src)>;
1965 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
1966 (VCVTTPD2DQYrr VR256:$src)>;
1967 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
1968 (VCVTTPD2DQYrm addr:$src)>;
1970 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
1971 (VCVTTPS2DQYrr VR256:$src)>;
1972 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
1973 (VCVTTPS2DQYrm addr:$src)>;
1975 // Match fround and fextend for 128/256-bit conversions
1976 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
1977 (VCVTPD2PSYrr VR256:$src)>;
1978 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
1979 (VCVTPD2PSYrm addr:$src)>;
1981 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
1982 (VCVTPS2PDYrr VR128:$src)>;
1983 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
1984 (VCVTPS2PDYrm addr:$src)>;
1986 //===----------------------------------------------------------------------===//
1987 // SSE 1 & 2 - Compare Instructions
1988 //===----------------------------------------------------------------------===//
1990 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
1991 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
1992 SDNode OpNode, ValueType VT, PatFrag ld_frag,
1993 string asm, string asm_alt> {
1994 def rr : SIi8<0xC2, MRMSrcReg,
1995 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
1996 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
1997 def rm : SIi8<0xC2, MRMSrcMem,
1998 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
1999 [(set RC:$dst, (OpNode (VT RC:$src1),
2000 (ld_frag addr:$src2), imm:$cc))]>;
2002 // Accept explicit immediate argument form instead of comparison code.
2003 let neverHasSideEffects = 1 in {
2004 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2005 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2007 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2008 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2012 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2013 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2014 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2015 XS, VEX_4V, VEX_LIG;
2016 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2017 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2018 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2019 XD, VEX_4V, VEX_LIG;
2021 let Constraints = "$src1 = $dst" in {
2022 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2023 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2024 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2026 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2027 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2028 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2032 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2033 Intrinsic Int, string asm> {
2034 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2035 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2036 [(set VR128:$dst, (Int VR128:$src1,
2037 VR128:$src, imm:$cc))]>;
2038 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2039 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2040 [(set VR128:$dst, (Int VR128:$src1,
2041 (load addr:$src), imm:$cc))]>;
2044 // Aliases to match intrinsics which expect XMM operand(s).
2045 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2046 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2048 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2049 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2051 let Constraints = "$src1 = $dst" in {
2052 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2053 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2054 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2055 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2059 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2060 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2061 ValueType vt, X86MemOperand x86memop,
2062 PatFrag ld_frag, string OpcodeStr, Domain d> {
2063 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2064 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2065 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2066 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2067 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2068 [(set EFLAGS, (OpNode (vt RC:$src1),
2069 (ld_frag addr:$src2)))], d>;
2072 let Defs = [EFLAGS] in {
2073 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2074 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2075 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2076 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2078 let Pattern = []<dag> in {
2079 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2080 "comiss", SSEPackedSingle>, TB, VEX,
2082 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2083 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2087 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2088 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2089 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2090 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2092 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2093 load, "comiss", SSEPackedSingle>, TB, VEX;
2094 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2095 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2096 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2097 "ucomiss", SSEPackedSingle>, TB;
2098 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2099 "ucomisd", SSEPackedDouble>, TB, OpSize;
2101 let Pattern = []<dag> in {
2102 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2103 "comiss", SSEPackedSingle>, TB;
2104 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2105 "comisd", SSEPackedDouble>, TB, OpSize;
2108 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2109 load, "ucomiss", SSEPackedSingle>, TB;
2110 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2111 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2113 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2114 "comiss", SSEPackedSingle>, TB;
2115 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2116 "comisd", SSEPackedDouble>, TB, OpSize;
2117 } // Defs = [EFLAGS]
2119 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2120 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2121 Intrinsic Int, string asm, string asm_alt,
2123 let isAsmParserOnly = 1 in {
2124 def rri : PIi8<0xC2, MRMSrcReg,
2125 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2126 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2127 def rmi : PIi8<0xC2, MRMSrcMem,
2128 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2129 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2132 // Accept explicit immediate argument form instead of comparison code.
2133 def rri_alt : PIi8<0xC2, MRMSrcReg,
2134 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2136 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2137 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2141 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2142 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2143 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2144 SSEPackedSingle>, TB, VEX_4V;
2145 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2146 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2147 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2148 SSEPackedDouble>, TB, OpSize, VEX_4V;
2149 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2150 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2151 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2152 SSEPackedSingle>, TB, VEX_4V;
2153 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2154 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2155 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2156 SSEPackedDouble>, TB, OpSize, VEX_4V;
2157 let Constraints = "$src1 = $dst" in {
2158 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2159 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2160 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2161 SSEPackedSingle>, TB;
2162 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2163 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2164 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2165 SSEPackedDouble>, TB, OpSize;
2168 let Predicates = [HasSSE1] in {
2169 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2170 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2171 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2172 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2175 let Predicates = [HasSSE2] in {
2176 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2177 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2178 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2179 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2182 let Predicates = [HasAVX] in {
2183 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2184 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2185 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2186 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2187 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2188 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2189 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2190 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2192 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2193 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2194 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2195 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2196 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2197 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2198 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2199 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2202 //===----------------------------------------------------------------------===//
2203 // SSE 1 & 2 - Shuffle Instructions
2204 //===----------------------------------------------------------------------===//
2206 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2207 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2208 ValueType vt, string asm, PatFrag mem_frag,
2209 Domain d, bit IsConvertibleToThreeAddress = 0> {
2210 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2211 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2212 [(set RC:$dst, (vt (shufp:$src3
2213 RC:$src1, (mem_frag addr:$src2))))], d>;
2214 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2215 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2216 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2218 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2221 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2222 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2223 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2224 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2225 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2226 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2227 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2228 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2229 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2230 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2231 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2232 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2234 let Constraints = "$src1 = $dst" in {
2235 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2236 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2237 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2239 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2240 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2241 memopv2f64, SSEPackedDouble>, TB, OpSize;
2244 let Predicates = [HasSSE1] in {
2245 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2246 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2247 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2248 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2249 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2250 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2251 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2252 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2253 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2254 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2255 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2256 // fall back to this for SSE1)
2257 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2258 (SHUFPSrri VR128:$src2, VR128:$src1,
2259 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2260 // Special unary SHUFPSrri case.
2261 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2262 (SHUFPSrri VR128:$src1, VR128:$src1,
2263 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2266 let Predicates = [HasSSE2] in {
2267 // Special binary v4i32 shuffle cases with SHUFPS.
2268 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2269 (SHUFPSrri VR128:$src1, VR128:$src2,
2270 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2271 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2272 (bc_v4i32 (memopv2i64 addr:$src2)))),
2273 (SHUFPSrmi VR128:$src1, addr:$src2,
2274 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2275 // Special unary SHUFPDrri cases.
2276 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2277 (SHUFPDrri VR128:$src1, VR128:$src1,
2278 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2279 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2280 (SHUFPDrri VR128:$src1, VR128:$src1,
2281 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2282 // Special binary v2i64 shuffle cases using SHUFPDrri.
2283 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2284 (SHUFPDrri VR128:$src1, VR128:$src2,
2285 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2286 // Generic SHUFPD patterns
2287 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2288 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2289 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2290 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2291 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2292 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2293 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2296 let Predicates = [HasAVX] in {
2297 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2298 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2299 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2300 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2301 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2302 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2303 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2304 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2305 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2306 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2307 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2308 // fall back to this for SSE1)
2309 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2310 (VSHUFPSrri VR128:$src2, VR128:$src1,
2311 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2312 // Special unary SHUFPSrri case.
2313 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2314 (VSHUFPSrri VR128:$src1, VR128:$src1,
2315 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2316 // Special binary v4i32 shuffle cases with SHUFPS.
2317 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2318 (VSHUFPSrri VR128:$src1, VR128:$src2,
2319 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2320 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2321 (bc_v4i32 (memopv2i64 addr:$src2)))),
2322 (VSHUFPSrmi VR128:$src1, addr:$src2,
2323 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2324 // Special unary SHUFPDrri cases.
2325 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2326 (VSHUFPDrri VR128:$src1, VR128:$src1,
2327 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2328 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2329 (VSHUFPDrri VR128:$src1, VR128:$src1,
2330 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2331 // Special binary v2i64 shuffle cases using SHUFPDrri.
2332 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2333 (VSHUFPDrri VR128:$src1, VR128:$src2,
2334 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2336 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2337 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2338 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2339 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2340 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2341 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2342 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2345 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2346 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2347 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2348 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2349 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2351 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2352 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2353 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2354 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2355 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2357 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2358 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2359 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2360 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2361 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2363 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2364 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2365 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2366 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2367 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2370 //===----------------------------------------------------------------------===//
2371 // SSE 1 & 2 - Unpack Instructions
2372 //===----------------------------------------------------------------------===//
2374 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2375 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2376 PatFrag mem_frag, RegisterClass RC,
2377 X86MemOperand x86memop, string asm,
2379 def rr : PI<opc, MRMSrcReg,
2380 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2382 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2383 def rm : PI<opc, MRMSrcMem,
2384 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2386 (vt (OpNode RC:$src1,
2387 (mem_frag addr:$src2))))], d>;
2390 let AddedComplexity = 10 in {
2391 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2392 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2393 SSEPackedSingle>, TB, VEX_4V;
2394 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2395 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2396 SSEPackedDouble>, TB, OpSize, VEX_4V;
2397 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2398 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2399 SSEPackedSingle>, TB, VEX_4V;
2400 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2401 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2402 SSEPackedDouble>, TB, OpSize, VEX_4V;
2404 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2405 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2406 SSEPackedSingle>, TB, VEX_4V;
2407 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2408 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2409 SSEPackedDouble>, TB, OpSize, VEX_4V;
2410 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2411 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2412 SSEPackedSingle>, TB, VEX_4V;
2413 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2414 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2415 SSEPackedDouble>, TB, OpSize, VEX_4V;
2417 let Constraints = "$src1 = $dst" in {
2418 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2419 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2420 SSEPackedSingle>, TB;
2421 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2422 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2423 SSEPackedDouble>, TB, OpSize;
2424 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2425 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2426 SSEPackedSingle>, TB;
2427 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2428 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2429 SSEPackedDouble>, TB, OpSize;
2430 } // Constraints = "$src1 = $dst"
2431 } // AddedComplexity
2433 let Predicates = [HasSSE1] in {
2434 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2435 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2436 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2437 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2438 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2439 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2440 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2441 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2444 let Predicates = [HasSSE2] in {
2445 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2446 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2447 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2448 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2449 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2450 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2451 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2452 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2454 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2455 // problem is during lowering, where it's not possible to recognize the load
2456 // fold cause it has two uses through a bitcast. One use disappears at isel
2457 // time and the fold opportunity reappears.
2458 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2459 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2461 let AddedComplexity = 10 in
2462 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2463 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2466 let Predicates = [HasAVX] in {
2467 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2468 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2469 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2470 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2471 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2472 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2473 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2474 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2476 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2477 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2478 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2479 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2480 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2481 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2482 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2483 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2485 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2486 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2487 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2488 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2489 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2490 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2491 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2492 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2494 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2495 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2496 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2497 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2498 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2499 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2500 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2501 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2503 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2504 // problem is during lowering, where it's not possible to recognize the load
2505 // fold cause it has two uses through a bitcast. One use disappears at isel
2506 // time and the fold opportunity reappears.
2507 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2508 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2509 let AddedComplexity = 10 in
2510 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2511 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2514 //===----------------------------------------------------------------------===//
2515 // SSE 1 & 2 - Extract Floating-Point Sign mask
2516 //===----------------------------------------------------------------------===//
2518 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2519 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2521 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2522 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2523 [(set GR32:$dst, (Int RC:$src))], d>;
2524 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2525 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2528 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2529 SSEPackedSingle>, TB;
2530 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2531 SSEPackedDouble>, TB, OpSize;
2533 def : Pat<(i32 (X86fgetsign FR32:$src)),
2534 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2535 sub_ss))>, Requires<[HasSSE1]>;
2536 def : Pat<(i64 (X86fgetsign FR32:$src)),
2537 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2538 sub_ss))>, Requires<[HasSSE1]>;
2539 def : Pat<(i32 (X86fgetsign FR64:$src)),
2540 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2541 sub_sd))>, Requires<[HasSSE2]>;
2542 def : Pat<(i64 (X86fgetsign FR64:$src)),
2543 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2544 sub_sd))>, Requires<[HasSSE2]>;
2546 let Predicates = [HasAVX] in {
2547 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2548 "movmskps", SSEPackedSingle>, TB, VEX;
2549 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2550 "movmskpd", SSEPackedDouble>, TB,
2552 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2553 "movmskps", SSEPackedSingle>, TB, VEX;
2554 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2555 "movmskpd", SSEPackedDouble>, TB,
2558 def : Pat<(i32 (X86fgetsign FR32:$src)),
2559 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2561 def : Pat<(i64 (X86fgetsign FR32:$src)),
2562 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2564 def : Pat<(i32 (X86fgetsign FR64:$src)),
2565 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2567 def : Pat<(i64 (X86fgetsign FR64:$src)),
2568 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2572 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2573 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2574 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2575 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2577 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2578 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2579 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2580 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2584 //===----------------------------------------------------------------------===//
2585 // SSE 1 & 2 - Logical Instructions
2586 //===----------------------------------------------------------------------===//
2588 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2590 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2592 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2593 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2595 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2596 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2598 let Constraints = "$src1 = $dst" in {
2599 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2600 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2602 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2603 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2607 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2608 let mayLoad = 0 in {
2609 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2610 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2611 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2614 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2615 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2617 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2619 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2621 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2622 // are all promoted to v2i64, and the patterns are covered by the int
2623 // version. This is needed in SSE only, because v2i64 isn't supported on
2624 // SSE1, but only on SSE2.
2625 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2626 !strconcat(OpcodeStr, "ps"), f128mem, [],
2627 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2628 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2630 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2631 !strconcat(OpcodeStr, "pd"), f128mem,
2632 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2633 (bc_v2i64 (v2f64 VR128:$src2))))],
2634 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2635 (memopv2i64 addr:$src2)))], 0>,
2637 let Constraints = "$src1 = $dst" in {
2638 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2639 !strconcat(OpcodeStr, "ps"), f128mem,
2640 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2641 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2642 (memopv2i64 addr:$src2)))]>, TB;
2644 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2645 !strconcat(OpcodeStr, "pd"), f128mem,
2646 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2647 (bc_v2i64 (v2f64 VR128:$src2))))],
2648 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2649 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2653 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2655 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2657 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2658 !strconcat(OpcodeStr, "ps"), f256mem,
2659 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2660 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2661 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2663 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2664 !strconcat(OpcodeStr, "pd"), f256mem,
2665 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2666 (bc_v4i64 (v4f64 VR256:$src2))))],
2667 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2668 (memopv4i64 addr:$src2)))], 0>,
2672 // AVX 256-bit packed logical ops forms
2673 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2674 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2675 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2676 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2678 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2679 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2680 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2681 let isCommutable = 0 in
2682 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2684 //===----------------------------------------------------------------------===//
2685 // SSE 1 & 2 - Arithmetic Instructions
2686 //===----------------------------------------------------------------------===//
2688 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2691 /// In addition, we also have a special variant of the scalar form here to
2692 /// represent the associated intrinsic operation. This form is unlike the
2693 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2694 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2696 /// These three forms can each be reg+reg or reg+mem.
2699 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2701 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2703 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2704 OpNode, FR32, f32mem, Is2Addr>, XS;
2705 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2706 OpNode, FR64, f64mem, Is2Addr>, XD;
2709 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2711 let mayLoad = 0 in {
2712 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2713 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2714 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2715 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2719 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2721 let mayLoad = 0 in {
2722 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2723 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2724 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2725 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2729 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2731 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2732 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2733 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2734 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2737 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2739 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2740 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2741 SSEPackedSingle, Is2Addr>, TB;
2743 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2744 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2745 SSEPackedDouble, Is2Addr>, TB, OpSize;
2748 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2749 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2750 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2751 SSEPackedSingle, 0>, TB;
2753 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2754 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2755 SSEPackedDouble, 0>, TB, OpSize;
2758 // Binary Arithmetic instructions
2759 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2760 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2761 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2762 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2763 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2764 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2765 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2766 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2768 let isCommutable = 0 in {
2769 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2770 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2771 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2772 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2773 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2774 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2775 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2776 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2777 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2778 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2779 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2780 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2781 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2782 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2783 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2784 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2785 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2786 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2787 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2788 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2791 let Constraints = "$src1 = $dst" in {
2792 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2793 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2794 basic_sse12_fp_binop_s_int<0x58, "add">;
2795 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2796 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2797 basic_sse12_fp_binop_s_int<0x59, "mul">;
2799 let isCommutable = 0 in {
2800 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2801 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2802 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2803 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2804 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2805 basic_sse12_fp_binop_s_int<0x5E, "div">;
2806 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2807 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2808 basic_sse12_fp_binop_s_int<0x5F, "max">,
2809 basic_sse12_fp_binop_p_int<0x5F, "max">;
2810 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2811 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2812 basic_sse12_fp_binop_s_int<0x5D, "min">,
2813 basic_sse12_fp_binop_p_int<0x5D, "min">;
2818 /// In addition, we also have a special variant of the scalar form here to
2819 /// represent the associated intrinsic operation. This form is unlike the
2820 /// plain scalar form, in that it takes an entire vector (instead of a
2821 /// scalar) and leaves the top elements undefined.
2823 /// And, we have a special variant form for a full-vector intrinsic form.
2825 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2826 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2827 SDNode OpNode, Intrinsic F32Int> {
2828 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2829 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2830 [(set FR32:$dst, (OpNode FR32:$src))]>;
2831 // For scalar unary operations, fold a load into the operation
2832 // only in OptForSize mode. It eliminates an instruction, but it also
2833 // eliminates a whole-register clobber (the load), so it introduces a
2834 // partial register update condition.
2835 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2836 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2837 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2838 Requires<[HasSSE1, OptForSize]>;
2839 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2840 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2841 [(set VR128:$dst, (F32Int VR128:$src))]>;
2842 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2843 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2844 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2847 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2848 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2849 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2850 !strconcat(OpcodeStr,
2851 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2853 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2854 !strconcat(OpcodeStr,
2855 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2856 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2857 (ins VR128:$src1, ssmem:$src2),
2858 !strconcat(OpcodeStr,
2859 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2862 /// sse1_fp_unop_p - SSE1 unops in packed form.
2863 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2864 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2865 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2866 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2867 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2868 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2869 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2872 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2873 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2874 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2875 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2876 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2877 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2878 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2879 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2882 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2883 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2884 Intrinsic V4F32Int> {
2885 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2886 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2887 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2888 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2889 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2890 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2893 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2894 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2895 Intrinsic V4F32Int> {
2896 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2897 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2898 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2899 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2900 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2901 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2904 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2905 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2906 SDNode OpNode, Intrinsic F64Int> {
2907 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2908 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2909 [(set FR64:$dst, (OpNode FR64:$src))]>;
2910 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2911 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2912 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2913 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2914 Requires<[HasSSE2, OptForSize]>;
2915 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2916 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2917 [(set VR128:$dst, (F64Int VR128:$src))]>;
2918 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2919 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2920 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2923 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2924 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2925 let neverHasSideEffects = 1 in {
2926 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2927 !strconcat(OpcodeStr,
2928 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2930 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2931 !strconcat(OpcodeStr,
2932 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2934 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2935 (ins VR128:$src1, sdmem:$src2),
2936 !strconcat(OpcodeStr,
2937 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2940 /// sse2_fp_unop_p - SSE2 unops in vector forms.
2941 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
2943 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2944 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2945 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
2946 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2947 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2948 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
2951 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
2952 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2953 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2954 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2955 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
2956 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2957 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2958 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
2961 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
2962 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2963 Intrinsic V2F64Int> {
2964 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2965 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2966 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
2967 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2968 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2969 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
2972 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
2973 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2974 Intrinsic V2F64Int> {
2975 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2976 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2977 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
2978 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2979 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
2980 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
2983 let Predicates = [HasAVX] in {
2985 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
2986 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
2988 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
2989 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
2990 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2991 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
2992 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
2993 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
2994 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
2995 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
2998 // Reciprocal approximations. Note that these typically require refinement
2999 // in order to obtain suitable precision.
3000 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3001 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3002 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3003 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3004 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3006 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3007 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3008 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3009 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3010 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3013 def : Pat<(f32 (fsqrt FR32:$src)),
3014 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3015 def : Pat<(f32 (fsqrt (load addr:$src))),
3016 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3017 Requires<[HasAVX, OptForSize]>;
3018 def : Pat<(f64 (fsqrt FR64:$src)),
3019 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3020 def : Pat<(f64 (fsqrt (load addr:$src))),
3021 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3022 Requires<[HasAVX, OptForSize]>;
3024 def : Pat<(f32 (X86frsqrt FR32:$src)),
3025 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3026 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3027 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3028 Requires<[HasAVX, OptForSize]>;
3030 def : Pat<(f32 (X86frcp FR32:$src)),
3031 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3032 def : Pat<(f32 (X86frcp (load addr:$src))),
3033 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3034 Requires<[HasAVX, OptForSize]>;
3036 let Predicates = [HasAVX] in {
3037 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3038 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3039 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3040 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3042 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3043 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3045 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3046 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3047 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3048 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3050 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3051 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3053 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3054 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3055 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3056 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3058 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3059 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3061 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3062 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3063 (VRCPSSr (f32 (IMPLICIT_DEF)),
3064 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3066 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3067 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3071 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3072 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3073 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3074 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3075 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3076 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3078 // Reciprocal approximations. Note that these typically require refinement
3079 // in order to obtain suitable precision.
3080 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3081 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3082 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3083 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3084 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3085 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3087 // There is no f64 version of the reciprocal approximation instructions.
3089 //===----------------------------------------------------------------------===//
3090 // SSE 1 & 2 - Non-temporal stores
3091 //===----------------------------------------------------------------------===//
3093 let AddedComplexity = 400 in { // Prefer non-temporal versions
3094 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3095 (ins f128mem:$dst, VR128:$src),
3096 "movntps\t{$src, $dst|$dst, $src}",
3097 [(alignednontemporalstore (v4f32 VR128:$src),
3099 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3100 (ins f128mem:$dst, VR128:$src),
3101 "movntpd\t{$src, $dst|$dst, $src}",
3102 [(alignednontemporalstore (v2f64 VR128:$src),
3104 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3105 (ins f128mem:$dst, VR128:$src),
3106 "movntdq\t{$src, $dst|$dst, $src}",
3107 [(alignednontemporalstore (v2f64 VR128:$src),
3110 let ExeDomain = SSEPackedInt in
3111 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3112 (ins f128mem:$dst, VR128:$src),
3113 "movntdq\t{$src, $dst|$dst, $src}",
3114 [(alignednontemporalstore (v4f32 VR128:$src),
3117 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3118 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3120 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3121 (ins f256mem:$dst, VR256:$src),
3122 "movntps\t{$src, $dst|$dst, $src}",
3123 [(alignednontemporalstore (v8f32 VR256:$src),
3125 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3126 (ins f256mem:$dst, VR256:$src),
3127 "movntpd\t{$src, $dst|$dst, $src}",
3128 [(alignednontemporalstore (v4f64 VR256:$src),
3130 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3131 (ins f256mem:$dst, VR256:$src),
3132 "movntdq\t{$src, $dst|$dst, $src}",
3133 [(alignednontemporalstore (v4f64 VR256:$src),
3135 let ExeDomain = SSEPackedInt in
3136 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3137 (ins f256mem:$dst, VR256:$src),
3138 "movntdq\t{$src, $dst|$dst, $src}",
3139 [(alignednontemporalstore (v8f32 VR256:$src),
3143 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3144 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3145 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3146 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3147 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3148 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3150 let AddedComplexity = 400 in { // Prefer non-temporal versions
3151 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3152 "movntps\t{$src, $dst|$dst, $src}",
3153 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3154 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3155 "movntpd\t{$src, $dst|$dst, $src}",
3156 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3158 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3159 "movntdq\t{$src, $dst|$dst, $src}",
3160 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3162 let ExeDomain = SSEPackedInt in
3163 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3164 "movntdq\t{$src, $dst|$dst, $src}",
3165 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3167 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3168 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3170 // There is no AVX form for instructions below this point
3171 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3172 "movnti{l}\t{$src, $dst|$dst, $src}",
3173 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3174 TB, Requires<[HasSSE2]>;
3175 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3176 "movnti{q}\t{$src, $dst|$dst, $src}",
3177 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3178 TB, Requires<[HasSSE2]>;
3181 //===----------------------------------------------------------------------===//
3182 // SSE 1 & 2 - Prefetch and memory fence
3183 //===----------------------------------------------------------------------===//
3185 // Prefetch intrinsic.
3186 def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3187 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3188 def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3189 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3190 def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3191 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3192 def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3193 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3196 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3197 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3198 TB, Requires<[HasSSE2]>;
3200 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3201 // was introduced with SSE2, it's backward compatible.
3202 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3204 // Load, store, and memory fence
3205 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3206 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3207 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3208 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3209 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3210 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3212 def : Pat<(X86SFence), (SFENCE)>;
3213 def : Pat<(X86LFence), (LFENCE)>;
3214 def : Pat<(X86MFence), (MFENCE)>;
3216 //===----------------------------------------------------------------------===//
3217 // SSE 1 & 2 - Load/Store XCSR register
3218 //===----------------------------------------------------------------------===//
3220 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3221 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3222 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3223 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3225 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3226 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3227 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3228 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3230 //===---------------------------------------------------------------------===//
3231 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3232 //===---------------------------------------------------------------------===//
3234 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3236 let neverHasSideEffects = 1 in {
3237 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3238 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3239 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3240 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3242 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3243 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3244 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3245 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3248 let isCodeGenOnly = 1 in {
3249 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3250 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3251 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3252 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3253 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3254 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3255 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3256 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3259 let canFoldAsLoad = 1, mayLoad = 1 in {
3260 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3261 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3262 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3263 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3264 let Predicates = [HasAVX] in {
3265 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3266 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3267 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3268 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3272 let mayStore = 1 in {
3273 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3274 (ins i128mem:$dst, VR128:$src),
3275 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3276 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3277 (ins i256mem:$dst, VR256:$src),
3278 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3279 let Predicates = [HasAVX] in {
3280 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3281 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3282 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3283 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3287 let neverHasSideEffects = 1 in
3288 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3289 "movdqa\t{$src, $dst|$dst, $src}", []>;
3291 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3292 "movdqu\t{$src, $dst|$dst, $src}",
3293 []>, XS, Requires<[HasSSE2]>;
3296 let isCodeGenOnly = 1 in {
3297 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3298 "movdqa\t{$src, $dst|$dst, $src}", []>;
3300 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3301 "movdqu\t{$src, $dst|$dst, $src}",
3302 []>, XS, Requires<[HasSSE2]>;
3305 let canFoldAsLoad = 1, mayLoad = 1 in {
3306 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3307 "movdqa\t{$src, $dst|$dst, $src}",
3308 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3309 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3310 "movdqu\t{$src, $dst|$dst, $src}",
3311 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3312 XS, Requires<[HasSSE2]>;
3315 let mayStore = 1 in {
3316 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3317 "movdqa\t{$src, $dst|$dst, $src}",
3318 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3319 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3320 "movdqu\t{$src, $dst|$dst, $src}",
3321 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3322 XS, Requires<[HasSSE2]>;
3325 // Intrinsic forms of MOVDQU load and store
3326 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3327 "vmovdqu\t{$src, $dst|$dst, $src}",
3328 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3329 XS, VEX, Requires<[HasAVX]>;
3331 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3332 "movdqu\t{$src, $dst|$dst, $src}",
3333 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3334 XS, Requires<[HasSSE2]>;
3336 } // ExeDomain = SSEPackedInt
3338 let Predicates = [HasAVX] in {
3339 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3340 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3341 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3344 //===---------------------------------------------------------------------===//
3345 // SSE2 - Packed Integer Arithmetic Instructions
3346 //===---------------------------------------------------------------------===//
3348 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3350 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3351 RegisterClass RC, PatFrag memop_frag,
3352 X86MemOperand x86memop, bit IsCommutable = 0,
3354 let isCommutable = IsCommutable in
3355 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3356 (ins RC:$src1, RC:$src2),
3358 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3359 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3360 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3361 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3362 (ins RC:$src1, x86memop:$src2),
3364 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3365 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3366 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3369 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3370 string OpcodeStr, Intrinsic IntId,
3371 Intrinsic IntId2, RegisterClass RC,
3373 // src2 is always 128-bit
3374 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3375 (ins RC:$src1, VR128:$src2),
3377 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3378 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3379 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3380 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3381 (ins RC:$src1, i128mem:$src2),
3383 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3384 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3385 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3386 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3387 (ins RC:$src1, i32i8imm:$src2),
3389 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3390 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3391 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3394 /// PDI_binop_rm - Simple SSE2 binary operator.
3395 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3396 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
3397 X86MemOperand x86memop, bit IsCommutable = 0,
3399 let isCommutable = IsCommutable in
3400 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3401 (ins RC:$src1, RC:$src2),
3403 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3404 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3405 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
3406 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3407 (ins RC:$src1, x86memop:$src2),
3409 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3410 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3411 [(set RC:$dst, (OpVT (OpNode RC:$src1,
3412 (bitconvert (memop_frag addr:$src2)))))]>;
3414 } // ExeDomain = SSEPackedInt
3416 // 128-bit Integer Arithmetic
3418 let Predicates = [HasAVX] in {
3419 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3420 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3421 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3422 i128mem, 1, 0>, VEX_4V;
3423 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3424 i128mem, 1, 0>, VEX_4V;
3425 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3426 i128mem, 1, 0>, VEX_4V;
3427 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3428 i128mem, 1, 0>, VEX_4V;
3429 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3430 i128mem, 0, 0>, VEX_4V;
3431 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3432 i128mem, 0, 0>, VEX_4V;
3433 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3434 i128mem, 0, 0>, VEX_4V;
3435 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3436 i128mem, 0, 0>, VEX_4V;
3439 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3440 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3441 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3442 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3443 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3444 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3445 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3446 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3447 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3448 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3449 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3450 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3451 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3452 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3453 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3454 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3455 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3456 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3457 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3458 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3459 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3460 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3461 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3462 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3463 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3464 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3465 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3466 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3467 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3468 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3469 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3470 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3471 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3472 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3473 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3474 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3475 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3476 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3479 let Predicates = [HasAVX2] in {
3480 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3481 i256mem, 1, 0>, VEX_4V;
3482 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3483 i256mem, 1, 0>, VEX_4V;
3484 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3485 i256mem, 1, 0>, VEX_4V;
3486 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3487 i256mem, 1, 0>, VEX_4V;
3488 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3489 i256mem, 1, 0>, VEX_4V;
3490 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3491 i256mem, 0, 0>, VEX_4V;
3492 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3493 i256mem, 0, 0>, VEX_4V;
3494 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3495 i256mem, 0, 0>, VEX_4V;
3496 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3497 i256mem, 0, 0>, VEX_4V;
3500 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3501 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3502 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3503 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3504 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3505 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3506 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3507 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3508 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3509 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3510 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3511 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3512 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3513 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3514 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3515 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3516 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3517 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3518 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3519 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3520 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3521 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3522 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3523 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3524 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3525 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3526 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3527 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3528 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3529 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3530 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3531 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3532 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3533 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3534 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3535 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3536 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3537 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3540 let Constraints = "$src1 = $dst" in {
3541 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3543 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3545 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3547 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3549 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3551 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3553 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3555 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3557 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3561 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3562 VR128, memopv2i64, i128mem>;
3563 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3564 VR128, memopv2i64, i128mem>;
3565 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3566 VR128, memopv2i64, i128mem>;
3567 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3568 VR128, memopv2i64, i128mem>;
3569 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3570 VR128, memopv2i64, i128mem, 1>;
3571 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3572 VR128, memopv2i64, i128mem, 1>;
3573 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3574 VR128, memopv2i64, i128mem, 1>;
3575 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3576 VR128, memopv2i64, i128mem, 1>;
3577 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3578 VR128, memopv2i64, i128mem, 1>;
3579 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3580 VR128, memopv2i64, i128mem, 1>;
3581 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3582 VR128, memopv2i64, i128mem, 1>;
3583 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3584 VR128, memopv2i64, i128mem, 1>;
3585 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3586 VR128, memopv2i64, i128mem, 1>;
3587 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3588 VR128, memopv2i64, i128mem, 1>;
3589 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3590 VR128, memopv2i64, i128mem, 1>;
3591 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3592 VR128, memopv2i64, i128mem, 1>;
3593 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3594 VR128, memopv2i64, i128mem, 1>;
3595 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3596 VR128, memopv2i64, i128mem, 1>;
3597 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3598 VR128, memopv2i64, i128mem, 1>;
3600 } // Constraints = "$src1 = $dst"
3602 //===---------------------------------------------------------------------===//
3603 // SSE2 - Packed Integer Logical Instructions
3604 //===---------------------------------------------------------------------===//
3606 let Predicates = [HasAVX] in {
3607 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3608 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3610 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3611 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3613 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3614 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3617 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3618 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3620 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3621 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3623 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3624 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3627 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3628 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3630 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3631 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3634 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
3635 i128mem, 1, 0>, VEX_4V;
3636 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
3637 i128mem, 1, 0>, VEX_4V;
3638 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
3639 i128mem, 1, 0>, VEX_4V;
3641 let ExeDomain = SSEPackedInt in {
3642 let neverHasSideEffects = 1 in {
3643 // 128-bit logical shifts.
3644 def VPSLLDQri : PDIi8<0x73, MRM7r,
3645 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3646 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3648 def VPSRLDQri : PDIi8<0x73, MRM3r,
3649 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3650 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3652 // PSRADQri doesn't exist in SSE[1-3].
3654 def VPANDNrr : PDI<0xDF, MRMSrcReg,
3655 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3656 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3658 (v2i64 (X86andnp VR128:$src1, VR128:$src2)))]>,VEX_4V;
3660 def VPANDNrm : PDI<0xDF, MRMSrcMem,
3661 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3662 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3663 [(set VR128:$dst, (X86andnp VR128:$src1,
3664 (memopv2i64 addr:$src2)))]>, VEX_4V;
3668 let Predicates = [HasAVX2] in {
3669 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3670 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3672 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3673 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3675 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3676 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3679 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3680 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3682 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3683 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3685 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3686 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3689 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3690 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3692 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3693 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3696 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
3697 i256mem, 1, 0>, VEX_4V;
3698 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
3699 i256mem, 1, 0>, VEX_4V;
3700 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
3701 i256mem, 1, 0>, VEX_4V;
3703 let ExeDomain = SSEPackedInt in {
3704 let neverHasSideEffects = 1 in {
3705 // 128-bit logical shifts.
3706 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3707 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3708 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3710 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3711 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3712 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3714 // PSRADQYri doesn't exist in SSE[1-3].
3716 def VPANDNYrr : PDI<0xDF, MRMSrcReg,
3717 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
3718 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3720 (v4i64 (X86andnp VR256:$src1, VR256:$src2)))]>,VEX_4V;
3722 def VPANDNYrm : PDI<0xDF, MRMSrcMem,
3723 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
3724 "vpandn\t{$src2, $src1, $dst|$dst, $src1, $src2}",
3725 [(set VR256:$dst, (X86andnp VR256:$src1,
3726 (memopv4i64 addr:$src2)))]>, VEX_4V;
3730 let Constraints = "$src1 = $dst" in {
3731 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3732 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3734 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3735 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3737 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3738 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3741 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3742 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3744 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3745 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3747 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3748 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3751 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3752 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3754 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3755 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3758 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
3760 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
3762 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
3765 let ExeDomain = SSEPackedInt in {
3766 let neverHasSideEffects = 1 in {
3767 // 128-bit logical shifts.
3768 def PSLLDQri : PDIi8<0x73, MRM7r,
3769 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3770 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3771 def PSRLDQri : PDIi8<0x73, MRM3r,
3772 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3773 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3774 // PSRADQri doesn't exist in SSE[1-3].
3775 def PANDNrr : PDI<0xDF, MRMSrcReg,
3776 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
3777 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3780 def PANDNrm : PDI<0xDF, MRMSrcMem,
3781 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
3782 "pandn\t{$src2, $dst|$dst, $src2}", []>;
3785 } // Constraints = "$src1 = $dst"
3787 let Predicates = [HasAVX] in {
3788 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3789 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3790 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3791 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3792 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3793 (VPSLLDQri VR128:$src1, imm:$src2)>;
3794 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3795 (VPSRLDQri VR128:$src1, imm:$src2)>;
3796 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3797 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3799 // Shift up / down and insert zero's.
3800 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3801 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3802 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3803 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3806 let Predicates = [HasAVX2] in {
3807 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3808 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3809 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3810 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3811 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3812 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3813 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3814 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3817 let Predicates = [HasSSE2] in {
3818 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3819 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3820 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3821 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3822 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3823 (PSLLDQri VR128:$src1, imm:$src2)>;
3824 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3825 (PSRLDQri VR128:$src1, imm:$src2)>;
3826 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3827 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3829 // Shift up / down and insert zero's.
3830 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3831 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3832 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3833 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3836 //===---------------------------------------------------------------------===//
3837 // SSE2 - Packed Integer Comparison Instructions
3838 //===---------------------------------------------------------------------===//
3840 let Predicates = [HasAVX] in {
3841 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3842 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3843 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3844 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3845 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3846 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3847 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3848 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3849 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3850 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3851 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3852 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3854 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3855 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3856 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3857 (bc_v16i8 (memopv2i64 addr:$src2)))),
3858 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3859 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3860 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3861 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3862 (bc_v8i16 (memopv2i64 addr:$src2)))),
3863 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3864 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3865 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3866 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3867 (bc_v4i32 (memopv2i64 addr:$src2)))),
3868 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3870 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3871 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3872 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3873 (bc_v16i8 (memopv2i64 addr:$src2)))),
3874 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3875 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3876 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3877 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3878 (bc_v8i16 (memopv2i64 addr:$src2)))),
3879 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3880 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3881 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3882 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3883 (bc_v4i32 (memopv2i64 addr:$src2)))),
3884 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3887 let Predicates = [HasAVX2] in {
3888 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3889 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3890 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3891 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3892 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3893 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3894 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3895 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3896 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3897 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3898 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3899 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3901 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3902 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3903 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3904 (bc_v32i8 (memopv4i64 addr:$src2)))),
3905 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3906 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3907 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3908 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3909 (bc_v16i16 (memopv4i64 addr:$src2)))),
3910 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3911 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3912 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3913 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3914 (bc_v8i32 (memopv4i64 addr:$src2)))),
3915 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3917 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3918 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3919 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3920 (bc_v32i8 (memopv4i64 addr:$src2)))),
3921 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3922 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3923 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3924 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3925 (bc_v16i16 (memopv4i64 addr:$src2)))),
3926 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3927 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3928 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3929 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3930 (bc_v8i32 (memopv4i64 addr:$src2)))),
3931 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3934 let Constraints = "$src1 = $dst" in {
3935 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3936 VR128, memopv2i64, i128mem, 1>;
3937 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3938 VR128, memopv2i64, i128mem, 1>;
3939 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3940 VR128, memopv2i64, i128mem, 1>;
3941 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3942 VR128, memopv2i64, i128mem>;
3943 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3944 VR128, memopv2i64, i128mem>;
3945 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3946 VR128, memopv2i64, i128mem>;
3947 } // Constraints = "$src1 = $dst"
3949 let Predicates = [HasSSE2] in {
3950 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3951 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3952 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3953 (bc_v16i8 (memopv2i64 addr:$src2)))),
3954 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3955 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3956 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3957 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3958 (bc_v8i16 (memopv2i64 addr:$src2)))),
3959 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3960 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3961 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3962 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3963 (bc_v4i32 (memopv2i64 addr:$src2)))),
3964 (PCMPEQDrm VR128:$src1, addr:$src2)>;
3966 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3967 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
3968 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3969 (bc_v16i8 (memopv2i64 addr:$src2)))),
3970 (PCMPGTBrm VR128:$src1, addr:$src2)>;
3971 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3972 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
3973 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3974 (bc_v8i16 (memopv2i64 addr:$src2)))),
3975 (PCMPGTWrm VR128:$src1, addr:$src2)>;
3976 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3977 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
3978 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3979 (bc_v4i32 (memopv2i64 addr:$src2)))),
3980 (PCMPGTDrm VR128:$src1, addr:$src2)>;
3983 //===---------------------------------------------------------------------===//
3984 // SSE2 - Packed Integer Pack Instructions
3985 //===---------------------------------------------------------------------===//
3987 let Predicates = [HasAVX] in {
3988 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
3989 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3990 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
3991 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3992 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
3993 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3996 let Predicates = [HasAVX2] in {
3997 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
3998 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3999 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4000 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4001 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4002 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4005 let Constraints = "$src1 = $dst" in {
4006 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4007 VR128, memopv2i64, i128mem>;
4008 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4009 VR128, memopv2i64, i128mem>;
4010 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4011 VR128, memopv2i64, i128mem>;
4012 } // Constraints = "$src1 = $dst"
4014 //===---------------------------------------------------------------------===//
4015 // SSE2 - Packed Integer Shuffle Instructions
4016 //===---------------------------------------------------------------------===//
4018 let ExeDomain = SSEPackedInt in {
4019 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4021 def ri : Ii8<0x70, MRMSrcReg,
4022 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4023 !strconcat(OpcodeStr,
4024 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4025 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4027 def mi : Ii8<0x70, MRMSrcMem,
4028 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4029 !strconcat(OpcodeStr,
4030 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4031 [(set VR128:$dst, (vt (pshuf_frag:$src2
4032 (bc_frag (memopv2i64 addr:$src1)),
4036 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4038 def Yri : Ii8<0x70, MRMSrcReg,
4039 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4040 !strconcat(OpcodeStr,
4041 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4042 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4044 def Ymi : Ii8<0x70, MRMSrcMem,
4045 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4046 !strconcat(OpcodeStr,
4047 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4048 [(set VR256:$dst, (vt (pshuf_frag:$src2
4049 (bc_frag (memopv4i64 addr:$src1)),
4052 } // ExeDomain = SSEPackedInt
4054 let Predicates = [HasAVX] in {
4055 let AddedComplexity = 5 in
4056 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4059 // SSE2 with ImmT == Imm8 and XS prefix.
4060 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4063 // SSE2 with ImmT == Imm8 and XD prefix.
4064 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4067 let AddedComplexity = 5 in
4068 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4069 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4070 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4071 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4072 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4074 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4076 (VPSHUFDmi addr:$src1, imm:$imm)>;
4077 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4079 (VPSHUFDmi addr:$src1, imm:$imm)>;
4080 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4081 (VPSHUFDri VR128:$src1, imm:$imm)>;
4082 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4083 (VPSHUFDri VR128:$src1, imm:$imm)>;
4084 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4085 (VPSHUFHWri VR128:$src, imm:$imm)>;
4086 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4088 (VPSHUFHWmi addr:$src, imm:$imm)>;
4089 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4090 (VPSHUFLWri VR128:$src, imm:$imm)>;
4091 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4093 (VPSHUFLWmi addr:$src, imm:$imm)>;
4096 let Predicates = [HasAVX2] in {
4097 let AddedComplexity = 5 in
4098 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4101 // SSE2 with ImmT == Imm8 and XS prefix.
4102 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4105 // SSE2 with ImmT == Imm8 and XD prefix.
4106 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4110 let Predicates = [HasSSE2] in {
4111 let AddedComplexity = 5 in
4112 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4114 // SSE2 with ImmT == Imm8 and XS prefix.
4115 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4117 // SSE2 with ImmT == Imm8 and XD prefix.
4118 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4120 let AddedComplexity = 5 in
4121 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4122 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4123 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4124 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4125 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4127 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4129 (PSHUFDmi addr:$src1, imm:$imm)>;
4130 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4132 (PSHUFDmi addr:$src1, imm:$imm)>;
4133 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4134 (PSHUFDri VR128:$src1, imm:$imm)>;
4135 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4136 (PSHUFDri VR128:$src1, imm:$imm)>;
4137 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4138 (PSHUFHWri VR128:$src, imm:$imm)>;
4139 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4141 (PSHUFHWmi addr:$src, imm:$imm)>;
4142 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4143 (PSHUFLWri VR128:$src, imm:$imm)>;
4144 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4146 (PSHUFLWmi addr:$src, imm:$imm)>;
4149 //===---------------------------------------------------------------------===//
4150 // SSE2 - Packed Integer Unpack Instructions
4151 //===---------------------------------------------------------------------===//
4153 let ExeDomain = SSEPackedInt in {
4154 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4155 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4156 def rr : PDI<opc, MRMSrcReg,
4157 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4159 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4160 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4161 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4162 def rm : PDI<opc, MRMSrcMem,
4163 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4165 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4166 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4167 [(set VR128:$dst, (OpNode VR128:$src1,
4168 (bc_frag (memopv2i64
4172 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4173 SDNode OpNode, PatFrag bc_frag> {
4174 def Yrr : PDI<opc, MRMSrcReg,
4175 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4176 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4177 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4178 def Yrm : PDI<opc, MRMSrcMem,
4179 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4180 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4181 [(set VR256:$dst, (OpNode VR256:$src1,
4182 (bc_frag (memopv4i64 addr:$src2))))]>;
4185 let Predicates = [HasAVX] in {
4186 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4187 bc_v16i8, 0>, VEX_4V;
4188 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4189 bc_v8i16, 0>, VEX_4V;
4190 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4191 bc_v4i32, 0>, VEX_4V;
4192 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4193 bc_v2i64, 0>, VEX_4V;
4195 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4196 bc_v16i8, 0>, VEX_4V;
4197 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4198 bc_v8i16, 0>, VEX_4V;
4199 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4200 bc_v4i32, 0>, VEX_4V;
4201 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4202 bc_v2i64, 0>, VEX_4V;
4205 let Predicates = [HasAVX2] in {
4206 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4208 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4210 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4212 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4215 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4217 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4219 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4221 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4225 let Constraints = "$src1 = $dst" in {
4226 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4228 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4230 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4232 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4235 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4237 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4239 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4241 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4244 } // ExeDomain = SSEPackedInt
4246 // Patterns for using AVX1 instructions with integer vectors
4247 // Here to give AVX2 priority
4248 let Predicates = [HasAVX] in {
4249 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4250 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4251 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4252 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4253 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4254 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4255 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4256 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4258 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4259 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4260 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4261 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4262 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4263 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4264 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4265 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4268 // Splat v2f64 / v2i64
4269 let AddedComplexity = 10 in {
4270 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4271 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4272 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4273 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4276 //===---------------------------------------------------------------------===//
4277 // SSE2 - Packed Integer Extract and Insert
4278 //===---------------------------------------------------------------------===//
4280 let ExeDomain = SSEPackedInt in {
4281 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4282 def rri : Ii8<0xC4, MRMSrcReg,
4283 (outs VR128:$dst), (ins VR128:$src1,
4284 GR32:$src2, i32i8imm:$src3),
4286 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4287 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4289 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4290 def rmi : Ii8<0xC4, MRMSrcMem,
4291 (outs VR128:$dst), (ins VR128:$src1,
4292 i16mem:$src2, i32i8imm:$src3),
4294 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4295 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4297 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4302 let Predicates = [HasAVX] in
4303 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4304 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4305 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4306 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4307 imm:$src2))]>, TB, OpSize, VEX;
4308 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4309 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4310 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4311 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4315 let Predicates = [HasAVX] in {
4316 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4317 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4318 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4319 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4320 []>, TB, OpSize, VEX_4V;
4323 let Constraints = "$src1 = $dst" in
4324 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4326 } // ExeDomain = SSEPackedInt
4328 //===---------------------------------------------------------------------===//
4329 // SSE2 - Packed Mask Creation
4330 //===---------------------------------------------------------------------===//
4332 let ExeDomain = SSEPackedInt in {
4334 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4335 "pmovmskb\t{$src, $dst|$dst, $src}",
4336 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4337 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4338 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4340 let Predicates = [HasAVX2] in {
4341 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4342 "pmovmskb\t{$src, $dst|$dst, $src}",
4343 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4344 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4345 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4348 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4349 "pmovmskb\t{$src, $dst|$dst, $src}",
4350 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4352 } // ExeDomain = SSEPackedInt
4354 //===---------------------------------------------------------------------===//
4355 // SSE2 - Conditional Store
4356 //===---------------------------------------------------------------------===//
4358 let ExeDomain = SSEPackedInt in {
4361 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4362 (ins VR128:$src, VR128:$mask),
4363 "maskmovdqu\t{$mask, $src|$src, $mask}",
4364 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4366 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4367 (ins VR128:$src, VR128:$mask),
4368 "maskmovdqu\t{$mask, $src|$src, $mask}",
4369 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4372 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4373 "maskmovdqu\t{$mask, $src|$src, $mask}",
4374 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4376 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4377 "maskmovdqu\t{$mask, $src|$src, $mask}",
4378 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4380 } // ExeDomain = SSEPackedInt
4382 //===---------------------------------------------------------------------===//
4383 // SSE2 - Move Doubleword
4384 //===---------------------------------------------------------------------===//
4386 //===---------------------------------------------------------------------===//
4387 // Move Int Doubleword to Packed Double Int
4389 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4390 "movd\t{$src, $dst|$dst, $src}",
4392 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4393 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4394 "movd\t{$src, $dst|$dst, $src}",
4396 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4398 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4399 "mov{d|q}\t{$src, $dst|$dst, $src}",
4401 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4402 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4403 "mov{d|q}\t{$src, $dst|$dst, $src}",
4404 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4406 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4407 "movd\t{$src, $dst|$dst, $src}",
4409 (v4i32 (scalar_to_vector GR32:$src)))]>;
4410 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4411 "movd\t{$src, $dst|$dst, $src}",
4413 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4414 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4415 "mov{d|q}\t{$src, $dst|$dst, $src}",
4417 (v2i64 (scalar_to_vector GR64:$src)))]>;
4418 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4419 "mov{d|q}\t{$src, $dst|$dst, $src}",
4420 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4422 //===---------------------------------------------------------------------===//
4423 // Move Int Doubleword to Single Scalar
4425 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4426 "movd\t{$src, $dst|$dst, $src}",
4427 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4429 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4430 "movd\t{$src, $dst|$dst, $src}",
4431 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4433 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4434 "movd\t{$src, $dst|$dst, $src}",
4435 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4437 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4438 "movd\t{$src, $dst|$dst, $src}",
4439 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4441 //===---------------------------------------------------------------------===//
4442 // Move Packed Doubleword Int to Packed Double Int
4444 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4445 "movd\t{$src, $dst|$dst, $src}",
4446 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4448 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4449 (ins i32mem:$dst, VR128:$src),
4450 "movd\t{$src, $dst|$dst, $src}",
4451 [(store (i32 (vector_extract (v4i32 VR128:$src),
4452 (iPTR 0))), addr:$dst)]>, VEX;
4453 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4454 "movd\t{$src, $dst|$dst, $src}",
4455 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4457 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4458 "movd\t{$src, $dst|$dst, $src}",
4459 [(store (i32 (vector_extract (v4i32 VR128:$src),
4460 (iPTR 0))), addr:$dst)]>;
4462 //===---------------------------------------------------------------------===//
4463 // Move Packed Doubleword Int first element to Doubleword Int
4465 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4466 "mov{d|q}\t{$src, $dst|$dst, $src}",
4467 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4469 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4471 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4472 "mov{d|q}\t{$src, $dst|$dst, $src}",
4473 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4476 //===---------------------------------------------------------------------===//
4477 // Bitcast FR64 <-> GR64
4479 let Predicates = [HasAVX] in
4480 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4481 "vmovq\t{$src, $dst|$dst, $src}",
4482 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4484 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4485 "mov{d|q}\t{$src, $dst|$dst, $src}",
4486 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4487 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4488 "movq\t{$src, $dst|$dst, $src}",
4489 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4491 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4492 "movq\t{$src, $dst|$dst, $src}",
4493 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4494 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4495 "mov{d|q}\t{$src, $dst|$dst, $src}",
4496 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4497 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4498 "movq\t{$src, $dst|$dst, $src}",
4499 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4501 //===---------------------------------------------------------------------===//
4502 // Move Scalar Single to Double Int
4504 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4505 "movd\t{$src, $dst|$dst, $src}",
4506 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4507 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4508 "movd\t{$src, $dst|$dst, $src}",
4509 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4510 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4511 "movd\t{$src, $dst|$dst, $src}",
4512 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4513 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4514 "movd\t{$src, $dst|$dst, $src}",
4515 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4517 //===---------------------------------------------------------------------===//
4518 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4520 let AddedComplexity = 15 in {
4521 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4522 "movd\t{$src, $dst|$dst, $src}",
4523 [(set VR128:$dst, (v4i32 (X86vzmovl
4524 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4526 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4527 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4528 [(set VR128:$dst, (v2i64 (X86vzmovl
4529 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4532 let AddedComplexity = 15 in {
4533 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4534 "movd\t{$src, $dst|$dst, $src}",
4535 [(set VR128:$dst, (v4i32 (X86vzmovl
4536 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4537 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4538 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4539 [(set VR128:$dst, (v2i64 (X86vzmovl
4540 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4543 let AddedComplexity = 20 in {
4544 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4545 "movd\t{$src, $dst|$dst, $src}",
4547 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4548 (loadi32 addr:$src))))))]>,
4550 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4551 "movd\t{$src, $dst|$dst, $src}",
4553 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4554 (loadi32 addr:$src))))))]>;
4557 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4558 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4559 (MOVZDI2PDIrm addr:$src)>;
4560 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4561 (MOVZDI2PDIrm addr:$src)>;
4562 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4563 (MOVZDI2PDIrm addr:$src)>;
4566 let Predicates = [HasAVX] in {
4567 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4568 let AddedComplexity = 20 in {
4569 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4570 (VMOVZDI2PDIrm addr:$src)>;
4571 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4572 (VMOVZDI2PDIrm addr:$src)>;
4573 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4574 (VMOVZDI2PDIrm addr:$src)>;
4576 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4577 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4578 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4579 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4580 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4581 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4582 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4585 // These are the correct encodings of the instructions so that we know how to
4586 // read correct assembly, even though we continue to emit the wrong ones for
4587 // compatibility with Darwin's buggy assembler.
4588 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4589 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4590 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4591 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4592 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4593 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4594 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4595 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4596 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4597 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4598 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4599 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4601 //===---------------------------------------------------------------------===//
4602 // SSE2 - Move Quadword
4603 //===---------------------------------------------------------------------===//
4605 //===---------------------------------------------------------------------===//
4606 // Move Quadword Int to Packed Quadword Int
4608 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4609 "vmovq\t{$src, $dst|$dst, $src}",
4611 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4612 VEX, Requires<[HasAVX]>;
4613 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4614 "movq\t{$src, $dst|$dst, $src}",
4616 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4617 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4619 //===---------------------------------------------------------------------===//
4620 // Move Packed Quadword Int to Quadword Int
4622 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4623 "movq\t{$src, $dst|$dst, $src}",
4624 [(store (i64 (vector_extract (v2i64 VR128:$src),
4625 (iPTR 0))), addr:$dst)]>, VEX;
4626 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4627 "movq\t{$src, $dst|$dst, $src}",
4628 [(store (i64 (vector_extract (v2i64 VR128:$src),
4629 (iPTR 0))), addr:$dst)]>;
4631 //===---------------------------------------------------------------------===//
4632 // Store / copy lower 64-bits of a XMM register.
4634 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4635 "movq\t{$src, $dst|$dst, $src}",
4636 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4637 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4638 "movq\t{$src, $dst|$dst, $src}",
4639 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4641 let AddedComplexity = 20 in
4642 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4643 "vmovq\t{$src, $dst|$dst, $src}",
4645 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4646 (loadi64 addr:$src))))))]>,
4647 XS, VEX, Requires<[HasAVX]>;
4649 let AddedComplexity = 20 in
4650 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4651 "movq\t{$src, $dst|$dst, $src}",
4653 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4654 (loadi64 addr:$src))))))]>,
4655 XS, Requires<[HasSSE2]>;
4657 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4658 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4659 (MOVZQI2PQIrm addr:$src)>;
4660 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4661 (MOVZQI2PQIrm addr:$src)>;
4662 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4665 let Predicates = [HasAVX], AddedComplexity = 20 in {
4666 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4667 (VMOVZQI2PQIrm addr:$src)>;
4668 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4669 (VMOVZQI2PQIrm addr:$src)>;
4670 def : Pat<(v2i64 (X86vzload addr:$src)),
4671 (VMOVZQI2PQIrm addr:$src)>;
4674 //===---------------------------------------------------------------------===//
4675 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4676 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4678 let AddedComplexity = 15 in
4679 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4680 "vmovq\t{$src, $dst|$dst, $src}",
4681 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4682 XS, VEX, Requires<[HasAVX]>;
4683 let AddedComplexity = 15 in
4684 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4685 "movq\t{$src, $dst|$dst, $src}",
4686 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4687 XS, Requires<[HasSSE2]>;
4689 let AddedComplexity = 20 in
4690 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4691 "vmovq\t{$src, $dst|$dst, $src}",
4692 [(set VR128:$dst, (v2i64 (X86vzmovl
4693 (loadv2i64 addr:$src))))]>,
4694 XS, VEX, Requires<[HasAVX]>;
4695 let AddedComplexity = 20 in {
4696 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4697 "movq\t{$src, $dst|$dst, $src}",
4698 [(set VR128:$dst, (v2i64 (X86vzmovl
4699 (loadv2i64 addr:$src))))]>,
4700 XS, Requires<[HasSSE2]>;
4703 let AddedComplexity = 20 in {
4704 let Predicates = [HasSSE2] in {
4705 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4706 (MOVZPQILo2PQIrm addr:$src)>;
4707 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4708 (MOVZPQILo2PQIrr VR128:$src)>;
4710 let Predicates = [HasAVX] in {
4711 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4712 (VMOVZPQILo2PQIrm addr:$src)>;
4713 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4714 (VMOVZPQILo2PQIrr VR128:$src)>;
4718 // Instructions to match in the assembler
4719 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4720 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4721 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4722 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4723 // Recognize "movd" with GR64 destination, but encode as a "movq"
4724 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4725 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4727 // Instructions for the disassembler
4728 // xr = XMM register
4731 let Predicates = [HasAVX] in
4732 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4733 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4734 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4735 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4737 //===---------------------------------------------------------------------===//
4738 // SSE3 - Conversion Instructions
4739 //===---------------------------------------------------------------------===//
4741 // Convert Packed Double FP to Packed DW Integers
4742 let Predicates = [HasAVX] in {
4743 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4744 // register, but the same isn't true when using memory operands instead.
4745 // Provide other assembly rr and rm forms to address this explicitly.
4746 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4747 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4748 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4749 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4752 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4753 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4754 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4755 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4758 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4759 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4760 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4761 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4764 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4765 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4766 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4767 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4769 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4770 (VCVTPD2DQYrr VR256:$src)>;
4771 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4772 (VCVTPD2DQYrm addr:$src)>;
4774 // Convert Packed DW Integers to Packed Double FP
4775 let Predicates = [HasAVX] in {
4776 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4777 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4778 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4779 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4780 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4781 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4782 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4783 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4786 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4787 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4788 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4789 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4791 // AVX 256-bit register conversion intrinsics
4792 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4793 (VCVTDQ2PDYrr VR128:$src)>;
4794 def : Pat<(int_x86_avx_cvtdq2_pd_256 (memopv4i32 addr:$src)),
4795 (VCVTDQ2PDYrm addr:$src)>;
4797 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4798 (VCVTPD2DQYrr VR256:$src)>;
4799 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4800 (VCVTPD2DQYrm addr:$src)>;
4802 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4803 (VCVTDQ2PDYrr VR128:$src)>;
4804 def : Pat<(v4f64 (sint_to_fp (memopv4i32 addr:$src))),
4805 (VCVTDQ2PDYrm addr:$src)>;
4807 //===---------------------------------------------------------------------===//
4808 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4809 //===---------------------------------------------------------------------===//
4810 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4811 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4812 X86MemOperand x86memop> {
4813 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4814 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4815 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4816 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4817 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4818 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4821 let Predicates = [HasAVX] in {
4822 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4823 v4f32, VR128, memopv4f32, f128mem>, VEX;
4824 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4825 v4f32, VR128, memopv4f32, f128mem>, VEX;
4826 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4827 v8f32, VR256, memopv8f32, f256mem>, VEX;
4828 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4829 v8f32, VR256, memopv8f32, f256mem>, VEX;
4831 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4832 memopv4f32, f128mem>;
4833 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4834 memopv4f32, f128mem>;
4836 let Predicates = [HasSSE3] in {
4837 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4838 (MOVSHDUPrr VR128:$src)>;
4839 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4840 (MOVSHDUPrm addr:$src)>;
4841 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4842 (MOVSLDUPrr VR128:$src)>;
4843 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4844 (MOVSLDUPrm addr:$src)>;
4847 let Predicates = [HasAVX] in {
4848 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4849 (VMOVSHDUPrr VR128:$src)>;
4850 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4851 (VMOVSHDUPrm addr:$src)>;
4852 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4853 (VMOVSLDUPrr VR128:$src)>;
4854 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4855 (VMOVSLDUPrm addr:$src)>;
4856 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4857 (VMOVSHDUPYrr VR256:$src)>;
4858 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4859 (VMOVSHDUPYrm addr:$src)>;
4860 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4861 (VMOVSLDUPYrr VR256:$src)>;
4862 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4863 (VMOVSLDUPYrm addr:$src)>;
4866 //===---------------------------------------------------------------------===//
4867 // SSE3 - Replicate Double FP - MOVDDUP
4868 //===---------------------------------------------------------------------===//
4870 multiclass sse3_replicate_dfp<string OpcodeStr> {
4871 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4872 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4873 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4874 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4875 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4877 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4881 // FIXME: Merge with above classe when there're patterns for the ymm version
4882 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4883 let Predicates = [HasAVX] in {
4884 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4885 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4887 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4888 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4893 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4894 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4895 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4897 let Predicates = [HasSSE3] in {
4898 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4900 (MOVDDUPrm addr:$src)>;
4901 let AddedComplexity = 5 in {
4902 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4903 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4904 (MOVDDUPrm addr:$src)>;
4905 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4906 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4907 (MOVDDUPrm addr:$src)>;
4909 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4910 (MOVDDUPrm addr:$src)>;
4911 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4912 (MOVDDUPrm addr:$src)>;
4913 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4914 (MOVDDUPrm addr:$src)>;
4915 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4916 (MOVDDUPrm addr:$src)>;
4917 def : Pat<(X86Movddup (bc_v2f64
4918 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4919 (MOVDDUPrm addr:$src)>;
4922 let Predicates = [HasAVX] in {
4923 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4925 (VMOVDDUPrm addr:$src)>;
4926 let AddedComplexity = 5 in {
4927 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4928 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4929 (VMOVDDUPrm addr:$src)>;
4930 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4931 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4932 (VMOVDDUPrm addr:$src)>;
4934 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4935 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4936 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4937 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4938 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4939 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4940 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4941 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4942 def : Pat<(X86Movddup (bc_v2f64
4943 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4944 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4947 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4948 (VMOVDDUPYrm addr:$src)>;
4949 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4950 (VMOVDDUPYrm addr:$src)>;
4951 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4952 (VMOVDDUPYrm addr:$src)>;
4953 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4954 (VMOVDDUPYrm addr:$src)>;
4955 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4956 (VMOVDDUPYrr VR256:$src)>;
4957 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4958 (VMOVDDUPYrr VR256:$src)>;
4961 //===---------------------------------------------------------------------===//
4962 // SSE3 - Move Unaligned Integer
4963 //===---------------------------------------------------------------------===//
4965 let Predicates = [HasAVX] in {
4966 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4967 "vlddqu\t{$src, $dst|$dst, $src}",
4968 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
4969 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
4970 "vlddqu\t{$src, $dst|$dst, $src}",
4971 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
4973 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4974 "lddqu\t{$src, $dst|$dst, $src}",
4975 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
4977 //===---------------------------------------------------------------------===//
4978 // SSE3 - Arithmetic
4979 //===---------------------------------------------------------------------===//
4981 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
4982 X86MemOperand x86memop, bit Is2Addr = 1> {
4983 def rr : I<0xD0, MRMSrcReg,
4984 (outs RC:$dst), (ins RC:$src1, RC:$src2),
4986 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4987 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4988 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
4989 def rm : I<0xD0, MRMSrcMem,
4990 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
4992 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
4993 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4994 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
4997 let Predicates = [HasAVX] in {
4998 let ExeDomain = SSEPackedSingle in {
4999 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5000 f128mem, 0>, TB, XD, VEX_4V;
5001 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5002 f256mem, 0>, TB, XD, VEX_4V;
5004 let ExeDomain = SSEPackedDouble in {
5005 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5006 f128mem, 0>, TB, OpSize, VEX_4V;
5007 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5008 f256mem, 0>, TB, OpSize, VEX_4V;
5011 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5012 let ExeDomain = SSEPackedSingle in
5013 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5015 let ExeDomain = SSEPackedDouble in
5016 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5017 f128mem>, TB, OpSize;
5020 //===---------------------------------------------------------------------===//
5021 // SSE3 Instructions
5022 //===---------------------------------------------------------------------===//
5025 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5026 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5027 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5029 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5030 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5031 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5033 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5035 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5036 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5037 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5039 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5040 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5041 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5043 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5044 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5045 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5047 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5049 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5050 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5051 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5054 let Predicates = [HasAVX] in {
5055 let ExeDomain = SSEPackedSingle in {
5056 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5057 X86fhadd, 0>, VEX_4V;
5058 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5059 X86fhsub, 0>, VEX_4V;
5060 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5061 X86fhadd, 0>, VEX_4V;
5062 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5063 X86fhsub, 0>, VEX_4V;
5065 let ExeDomain = SSEPackedDouble in {
5066 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5067 X86fhadd, 0>, VEX_4V;
5068 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5069 X86fhsub, 0>, VEX_4V;
5070 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5071 X86fhadd, 0>, VEX_4V;
5072 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5073 X86fhsub, 0>, VEX_4V;
5077 let Constraints = "$src1 = $dst" in {
5078 let ExeDomain = SSEPackedSingle in {
5079 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5080 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5082 let ExeDomain = SSEPackedDouble in {
5083 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5084 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5088 //===---------------------------------------------------------------------===//
5089 // SSSE3 - Packed Absolute Instructions
5090 //===---------------------------------------------------------------------===//
5093 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5094 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5095 PatFrag mem_frag128, Intrinsic IntId128> {
5096 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5098 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5099 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5102 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5104 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5107 (bitconvert (mem_frag128 addr:$src))))]>, OpSize;
5110 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5111 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5112 PatFrag mem_frag256, Intrinsic IntId256> {
5113 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5115 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5116 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5119 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5121 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5124 (bitconvert (mem_frag256 addr:$src))))]>, OpSize;
5127 let Predicates = [HasAVX] in {
5128 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb", memopv16i8,
5129 int_x86_ssse3_pabs_b_128>, VEX;
5130 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw", memopv8i16,
5131 int_x86_ssse3_pabs_w_128>, VEX;
5132 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd", memopv4i32,
5133 int_x86_ssse3_pabs_d_128>, VEX;
5136 let Predicates = [HasAVX2] in {
5137 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb", memopv32i8,
5138 int_x86_avx2_pabs_b>, VEX;
5139 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw", memopv16i16,
5140 int_x86_avx2_pabs_w>, VEX;
5141 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd", memopv8i32,
5142 int_x86_avx2_pabs_d>, VEX;
5145 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb", memopv16i8,
5146 int_x86_ssse3_pabs_b_128>;
5147 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw", memopv8i16,
5148 int_x86_ssse3_pabs_w_128>;
5149 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd", memopv4i32,
5150 int_x86_ssse3_pabs_d_128>;
5152 //===---------------------------------------------------------------------===//
5153 // SSSE3 - Packed Binary Operator Instructions
5154 //===---------------------------------------------------------------------===//
5156 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5157 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5158 PatFrag mem_frag128, Intrinsic IntId128,
5160 let isCommutable = 1 in
5161 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5162 (ins VR128:$src1, VR128:$src2),
5164 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5165 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5166 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5168 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5169 (ins VR128:$src1, i128mem:$src2),
5171 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5172 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5174 (IntId128 VR128:$src1,
5175 (bitconvert (mem_frag128 addr:$src2))))]>, OpSize;
5178 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5179 PatFrag mem_frag256, Intrinsic IntId256> {
5180 let isCommutable = 1 in
5181 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5182 (ins VR256:$src1, VR256:$src2),
5183 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5184 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5186 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5187 (ins VR256:$src1, i256mem:$src2),
5188 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5190 (IntId256 VR256:$src1,
5191 (bitconvert (mem_frag256 addr:$src2))))]>, OpSize;
5194 let ImmT = NoImm, Predicates = [HasAVX] in {
5195 let isCommutable = 0 in {
5196 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw", memopv8i16,
5197 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5198 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd", memopv4i32,
5199 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5200 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw", memopv8i16,
5201 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5202 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw", memopv8i16,
5203 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5204 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd", memopv4i32,
5205 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5206 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw", memopv8i16,
5207 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5208 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw", memopv16i8,
5209 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5210 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb", memopv16i8,
5211 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5212 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb", memopv16i8,
5213 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5214 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw", memopv8i16,
5215 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5216 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd", memopv4i32,
5217 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5219 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw", memopv8i16,
5220 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5223 let ImmT = NoImm, Predicates = [HasAVX2] in {
5224 let isCommutable = 0 in {
5225 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw", memopv16i16,
5226 int_x86_avx2_phadd_w>, VEX_4V;
5227 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd", memopv8i32,
5228 int_x86_avx2_phadd_d>, VEX_4V;
5229 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw", memopv16i16,
5230 int_x86_avx2_phadd_sw>, VEX_4V;
5231 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw", memopv16i16,
5232 int_x86_avx2_phsub_w>, VEX_4V;
5233 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd", memopv8i32,
5234 int_x86_avx2_phsub_d>, VEX_4V;
5235 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw", memopv16i16,
5236 int_x86_avx2_phsub_sw>, VEX_4V;
5237 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw", memopv32i8,
5238 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5239 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb", memopv32i8,
5240 int_x86_avx2_pshuf_b>, VEX_4V;
5241 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb", memopv32i8,
5242 int_x86_avx2_psign_b>, VEX_4V;
5243 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw", memopv16i16,
5244 int_x86_avx2_psign_w>, VEX_4V;
5245 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd", memopv8i32,
5246 int_x86_avx2_psign_d>, VEX_4V;
5248 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw", memopv16i16,
5249 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5252 // None of these have i8 immediate fields.
5253 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5254 let isCommutable = 0 in {
5255 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw", memopv8i16,
5256 int_x86_ssse3_phadd_w_128>;
5257 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd", memopv4i32,
5258 int_x86_ssse3_phadd_d_128>;
5259 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw", memopv8i16,
5260 int_x86_ssse3_phadd_sw_128>;
5261 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw", memopv8i16,
5262 int_x86_ssse3_phsub_w_128>;
5263 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd", memopv4i32,
5264 int_x86_ssse3_phsub_d_128>;
5265 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw", memopv8i16,
5266 int_x86_ssse3_phsub_sw_128>;
5267 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw", memopv16i8,
5268 int_x86_ssse3_pmadd_ub_sw_128>;
5269 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb", memopv16i8,
5270 int_x86_ssse3_pshuf_b_128>;
5271 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb", memopv16i8,
5272 int_x86_ssse3_psign_b_128>;
5273 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw", memopv8i16,
5274 int_x86_ssse3_psign_w_128>;
5275 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd", memopv4i32,
5276 int_x86_ssse3_psign_d_128>;
5278 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw", memopv8i16,
5279 int_x86_ssse3_pmul_hr_sw_128>;
5282 let Predicates = [HasSSSE3] in {
5283 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5284 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5285 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5286 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5288 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5289 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5290 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5291 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5292 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5293 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5295 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5296 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5297 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5298 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5299 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5300 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5301 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5302 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5305 let Predicates = [HasAVX] in {
5306 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5307 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5308 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5309 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5311 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5312 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5313 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5314 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5315 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5316 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5318 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5319 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5320 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5321 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5322 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5323 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5324 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5325 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5328 let Predicates = [HasAVX2] in {
5329 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5330 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5331 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5332 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5333 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5334 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5336 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5337 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5338 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5339 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5340 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5341 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5342 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5343 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5346 //===---------------------------------------------------------------------===//
5347 // SSSE3 - Packed Align Instruction Patterns
5348 //===---------------------------------------------------------------------===//
5350 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5351 let neverHasSideEffects = 1 in {
5352 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5353 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5355 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5357 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5360 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5361 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5363 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5365 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5370 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5371 let neverHasSideEffects = 1 in {
5372 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5373 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5375 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5378 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5379 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5381 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5386 let Predicates = [HasAVX] in
5387 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5388 let Predicates = [HasAVX2] in
5389 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5390 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5391 defm PALIGN : ssse3_palign<"palignr">;
5393 let Predicates = [HasSSSE3] in {
5394 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5395 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5396 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5397 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5398 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5399 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5400 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5401 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5404 let Predicates = [HasAVX] in {
5405 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5406 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5407 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5408 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5409 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5410 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5411 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5412 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5415 //===---------------------------------------------------------------------===//
5416 // SSSE3 - Thread synchronization
5417 //===---------------------------------------------------------------------===//
5419 let usesCustomInserter = 1 in {
5420 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5421 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5422 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5423 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5426 let Uses = [EAX, ECX, EDX] in
5427 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5428 Requires<[HasSSE3]>;
5429 let Uses = [ECX, EAX] in
5430 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5431 Requires<[HasSSE3]>;
5433 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5434 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5436 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5437 Requires<[In32BitMode]>;
5438 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5439 Requires<[In64BitMode]>;
5441 //===----------------------------------------------------------------------===//
5442 // SSE4.1 - Packed Move with Sign/Zero Extend
5443 //===----------------------------------------------------------------------===//
5445 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5446 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5447 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5448 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5450 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5451 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5453 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5457 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5459 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5460 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5461 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5463 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5464 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5465 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5468 let Predicates = [HasAVX] in {
5469 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5471 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5473 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5475 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5477 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5479 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5483 let Predicates = [HasAVX2] in {
5484 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5485 int_x86_avx2_pmovsxbw>, VEX;
5486 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5487 int_x86_avx2_pmovsxwd>, VEX;
5488 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5489 int_x86_avx2_pmovsxdq>, VEX;
5490 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5491 int_x86_avx2_pmovzxbw>, VEX;
5492 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5493 int_x86_avx2_pmovzxwd>, VEX;
5494 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5495 int_x86_avx2_pmovzxdq>, VEX;
5498 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5499 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5500 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5501 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5502 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5503 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5505 let Predicates = [HasSSE41] in {
5506 // Common patterns involving scalar load.
5507 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5508 (PMOVSXBWrm addr:$src)>;
5509 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5510 (PMOVSXBWrm addr:$src)>;
5512 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5513 (PMOVSXWDrm addr:$src)>;
5514 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5515 (PMOVSXWDrm addr:$src)>;
5517 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5518 (PMOVSXDQrm addr:$src)>;
5519 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5520 (PMOVSXDQrm addr:$src)>;
5522 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5523 (PMOVZXBWrm addr:$src)>;
5524 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5525 (PMOVZXBWrm addr:$src)>;
5527 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5528 (PMOVZXWDrm addr:$src)>;
5529 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5530 (PMOVZXWDrm addr:$src)>;
5532 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5533 (PMOVZXDQrm addr:$src)>;
5534 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5535 (PMOVZXDQrm addr:$src)>;
5538 let Predicates = [HasAVX] in {
5539 // Common patterns involving scalar load.
5540 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5541 (VPMOVSXBWrm addr:$src)>;
5542 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5543 (VPMOVSXBWrm addr:$src)>;
5545 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5546 (VPMOVSXWDrm addr:$src)>;
5547 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5548 (VPMOVSXWDrm addr:$src)>;
5550 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5551 (VPMOVSXDQrm addr:$src)>;
5552 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5553 (VPMOVSXDQrm addr:$src)>;
5555 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5556 (VPMOVZXBWrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5558 (VPMOVZXBWrm addr:$src)>;
5560 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5561 (VPMOVZXWDrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5563 (VPMOVZXWDrm addr:$src)>;
5565 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5566 (VPMOVZXDQrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5568 (VPMOVZXDQrm addr:$src)>;
5572 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5573 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5574 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5575 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5577 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5578 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5580 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5584 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5586 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5587 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5588 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5590 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5591 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5593 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5597 let Predicates = [HasAVX] in {
5598 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5600 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5602 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5604 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5608 let Predicates = [HasAVX2] in {
5609 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5610 int_x86_avx2_pmovsxbd>, VEX;
5611 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5612 int_x86_avx2_pmovsxwq>, VEX;
5613 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5614 int_x86_avx2_pmovzxbd>, VEX;
5615 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5616 int_x86_avx2_pmovzxwq>, VEX;
5619 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5620 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5621 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5622 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5624 let Predicates = [HasSSE41] in {
5625 // Common patterns involving scalar load
5626 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5627 (PMOVSXBDrm addr:$src)>;
5628 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5629 (PMOVSXWQrm addr:$src)>;
5631 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5632 (PMOVZXBDrm addr:$src)>;
5633 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5634 (PMOVZXWQrm addr:$src)>;
5637 let Predicates = [HasAVX] in {
5638 // Common patterns involving scalar load
5639 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5640 (VPMOVSXBDrm addr:$src)>;
5641 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5642 (VPMOVSXWQrm addr:$src)>;
5644 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5645 (VPMOVZXBDrm addr:$src)>;
5646 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5647 (VPMOVZXWQrm addr:$src)>;
5650 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5651 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5652 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5653 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5655 // Expecting a i16 load any extended to i32 value.
5656 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5657 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5658 [(set VR128:$dst, (IntId (bitconvert
5659 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5663 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5665 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5666 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5667 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5669 // Expecting a i16 load any extended to i32 value.
5670 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5671 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5672 [(set VR256:$dst, (IntId (bitconvert
5673 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5677 let Predicates = [HasAVX] in {
5678 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5680 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5683 let Predicates = [HasAVX2] in {
5684 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5685 int_x86_avx2_pmovsxbq>, VEX;
5686 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5687 int_x86_avx2_pmovzxbq>, VEX;
5689 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5690 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5692 let Predicates = [HasSSE41] in {
5693 // Common patterns involving scalar load
5694 def : Pat<(int_x86_sse41_pmovsxbq
5695 (bitconvert (v4i32 (X86vzmovl
5696 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5697 (PMOVSXBQrm addr:$src)>;
5699 def : Pat<(int_x86_sse41_pmovzxbq
5700 (bitconvert (v4i32 (X86vzmovl
5701 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5702 (PMOVZXBQrm addr:$src)>;
5705 let Predicates = [HasAVX] in {
5706 // Common patterns involving scalar load
5707 def : Pat<(int_x86_sse41_pmovsxbq
5708 (bitconvert (v4i32 (X86vzmovl
5709 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5710 (VPMOVSXBQrm addr:$src)>;
5712 def : Pat<(int_x86_sse41_pmovzxbq
5713 (bitconvert (v4i32 (X86vzmovl
5714 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5715 (VPMOVZXBQrm addr:$src)>;
5718 //===----------------------------------------------------------------------===//
5719 // SSE4.1 - Extract Instructions
5720 //===----------------------------------------------------------------------===//
5722 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5723 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5724 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5725 (ins VR128:$src1, i32i8imm:$src2),
5726 !strconcat(OpcodeStr,
5727 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5728 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5730 let neverHasSideEffects = 1, mayStore = 1 in
5731 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5732 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5733 !strconcat(OpcodeStr,
5734 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5737 // There's an AssertZext in the way of writing the store pattern
5738 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5741 let Predicates = [HasAVX] in {
5742 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5743 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5744 (ins VR128:$src1, i32i8imm:$src2),
5745 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5748 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5751 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5752 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5753 let neverHasSideEffects = 1, mayStore = 1 in
5754 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5755 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5756 !strconcat(OpcodeStr,
5757 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5760 // There's an AssertZext in the way of writing the store pattern
5761 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5764 let Predicates = [HasAVX] in
5765 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5767 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5770 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5771 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5772 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5773 (ins VR128:$src1, i32i8imm:$src2),
5774 !strconcat(OpcodeStr,
5775 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5777 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5778 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5779 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5780 !strconcat(OpcodeStr,
5781 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5782 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5783 addr:$dst)]>, OpSize;
5786 let Predicates = [HasAVX] in
5787 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5789 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5791 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5792 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5793 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5794 (ins VR128:$src1, i32i8imm:$src2),
5795 !strconcat(OpcodeStr,
5796 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5798 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5799 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5800 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5801 !strconcat(OpcodeStr,
5802 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5803 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5804 addr:$dst)]>, OpSize, REX_W;
5807 let Predicates = [HasAVX] in
5808 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5810 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5812 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5814 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5815 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5816 (ins VR128:$src1, i32i8imm:$src2),
5817 !strconcat(OpcodeStr,
5818 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5820 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5822 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5823 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5824 !strconcat(OpcodeStr,
5825 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5826 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5827 addr:$dst)]>, OpSize;
5830 let ExeDomain = SSEPackedSingle in {
5831 let Predicates = [HasAVX] in {
5832 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5833 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5834 (ins VR128:$src1, i32i8imm:$src2),
5835 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5838 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5841 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5842 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5845 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5846 Requires<[HasSSE41]>;
5847 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5850 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5853 //===----------------------------------------------------------------------===//
5854 // SSE4.1 - Insert Instructions
5855 //===----------------------------------------------------------------------===//
5857 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5858 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5859 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5861 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5863 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5865 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5866 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5867 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5869 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5871 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5873 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5874 imm:$src3))]>, OpSize;
5877 let Predicates = [HasAVX] in
5878 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5879 let Constraints = "$src1 = $dst" in
5880 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5882 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5883 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5884 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5886 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5888 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5890 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5892 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5893 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5895 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5897 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5899 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5900 imm:$src3)))]>, OpSize;
5903 let Predicates = [HasAVX] in
5904 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5905 let Constraints = "$src1 = $dst" in
5906 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5908 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5909 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5910 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5912 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5914 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5916 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5918 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5919 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5921 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5923 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5925 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5926 imm:$src3)))]>, OpSize;
5929 let Predicates = [HasAVX] in
5930 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5931 let Constraints = "$src1 = $dst" in
5932 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5934 // insertps has a few different modes, there's the first two here below which
5935 // are optimized inserts that won't zero arbitrary elements in the destination
5936 // vector. The next one matches the intrinsic and could zero arbitrary elements
5937 // in the target vector.
5938 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5939 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5940 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5942 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5944 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5946 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5948 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5949 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5951 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5953 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5955 (X86insrtps VR128:$src1,
5956 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5957 imm:$src3))]>, OpSize;
5960 let ExeDomain = SSEPackedSingle in {
5961 let Constraints = "$src1 = $dst" in
5962 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5963 let Predicates = [HasAVX] in
5964 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
5967 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5968 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5970 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
5971 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
5972 Requires<[HasSSE41]>;
5974 //===----------------------------------------------------------------------===//
5975 // SSE4.1 - Round Instructions
5976 //===----------------------------------------------------------------------===//
5978 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
5979 X86MemOperand x86memop, RegisterClass RC,
5980 PatFrag mem_frag32, PatFrag mem_frag64,
5981 Intrinsic V4F32Int, Intrinsic V2F64Int> {
5982 let ExeDomain = SSEPackedSingle in {
5983 // Intrinsic operation, reg.
5984 // Vector intrinsic operation, reg
5985 def PSr : SS4AIi8<opcps, MRMSrcReg,
5986 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
5987 !strconcat(OpcodeStr,
5988 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5989 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
5992 // Vector intrinsic operation, mem
5993 def PSm : SS4AIi8<opcps, MRMSrcMem,
5994 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
5995 !strconcat(OpcodeStr,
5996 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5998 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6000 } // ExeDomain = SSEPackedSingle
6002 let ExeDomain = SSEPackedDouble in {
6003 // Vector intrinsic operation, reg
6004 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6005 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6006 !strconcat(OpcodeStr,
6007 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6008 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6011 // Vector intrinsic operation, mem
6012 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6013 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6014 !strconcat(OpcodeStr,
6015 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6017 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6019 } // ExeDomain = SSEPackedDouble
6022 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6025 Intrinsic F64Int, bit Is2Addr = 1> {
6026 let ExeDomain = GenericDomain in {
6027 // Intrinsic operation, reg.
6028 def SSr : SS4AIi8<opcss, MRMSrcReg,
6029 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6031 !strconcat(OpcodeStr,
6032 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6033 !strconcat(OpcodeStr,
6034 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6035 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6038 // Intrinsic operation, mem.
6039 def SSm : SS4AIi8<opcss, MRMSrcMem,
6040 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6042 !strconcat(OpcodeStr,
6043 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6044 !strconcat(OpcodeStr,
6045 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6047 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6050 // Intrinsic operation, reg.
6051 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6052 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6054 !strconcat(OpcodeStr,
6055 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6056 !strconcat(OpcodeStr,
6057 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6058 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6061 // Intrinsic operation, mem.
6062 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6063 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6065 !strconcat(OpcodeStr,
6066 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6067 !strconcat(OpcodeStr,
6068 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6070 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6072 } // ExeDomain = GenericDomain
6075 // FP round - roundss, roundps, roundsd, roundpd
6076 let Predicates = [HasAVX] in {
6078 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6079 memopv4f32, memopv2f64,
6080 int_x86_sse41_round_ps,
6081 int_x86_sse41_round_pd>, VEX;
6082 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6083 memopv8f32, memopv4f64,
6084 int_x86_avx_round_ps_256,
6085 int_x86_avx_round_pd_256>, VEX;
6086 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6087 int_x86_sse41_round_ss,
6088 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6091 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6092 memopv4f32, memopv2f64,
6093 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6094 let Constraints = "$src1 = $dst" in
6095 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6096 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6098 //===----------------------------------------------------------------------===//
6099 // SSE4.1 - Packed Bit Test
6100 //===----------------------------------------------------------------------===//
6102 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6103 // the intel intrinsic that corresponds to this.
6104 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6105 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6106 "vptest\t{$src2, $src1|$src1, $src2}",
6107 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6109 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6110 "vptest\t{$src2, $src1|$src1, $src2}",
6111 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6114 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6115 "vptest\t{$src2, $src1|$src1, $src2}",
6116 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6118 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6119 "vptest\t{$src2, $src1|$src1, $src2}",
6120 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6124 let Defs = [EFLAGS] in {
6125 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6126 "ptest\t{$src2, $src1|$src1, $src2}",
6127 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6129 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6130 "ptest\t{$src2, $src1|$src1, $src2}",
6131 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6135 // The bit test instructions below are AVX only
6136 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6137 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6138 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6139 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6140 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6141 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6142 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6143 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6147 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6148 let ExeDomain = SSEPackedSingle in {
6149 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6150 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6152 let ExeDomain = SSEPackedDouble in {
6153 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6154 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6158 //===----------------------------------------------------------------------===//
6159 // SSE4.1 - Misc Instructions
6160 //===----------------------------------------------------------------------===//
6162 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6163 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6164 "popcnt{w}\t{$src, $dst|$dst, $src}",
6165 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6167 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6168 "popcnt{w}\t{$src, $dst|$dst, $src}",
6169 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6170 (implicit EFLAGS)]>, OpSize, XS;
6172 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6173 "popcnt{l}\t{$src, $dst|$dst, $src}",
6174 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6176 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6177 "popcnt{l}\t{$src, $dst|$dst, $src}",
6178 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6179 (implicit EFLAGS)]>, XS;
6181 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6182 "popcnt{q}\t{$src, $dst|$dst, $src}",
6183 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6185 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6186 "popcnt{q}\t{$src, $dst|$dst, $src}",
6187 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6188 (implicit EFLAGS)]>, XS;
6193 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6194 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6195 Intrinsic IntId128> {
6196 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6198 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6199 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6200 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6202 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6205 (bitconvert (memopv8i16 addr:$src))))]>, OpSize;
6208 let Predicates = [HasAVX] in
6209 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6210 int_x86_sse41_phminposuw>, VEX;
6211 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6212 int_x86_sse41_phminposuw>;
6214 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6215 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6216 Intrinsic IntId128, bit Is2Addr = 1> {
6217 let isCommutable = 1 in
6218 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6219 (ins VR128:$src1, VR128:$src2),
6221 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6222 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6223 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6224 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6225 (ins VR128:$src1, i128mem:$src2),
6227 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6228 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6230 (IntId128 VR128:$src1,
6231 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6234 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6235 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6236 Intrinsic IntId256> {
6237 let isCommutable = 1 in
6238 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6239 (ins VR256:$src1, VR256:$src2),
6240 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6241 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6242 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6243 (ins VR256:$src1, i256mem:$src2),
6244 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6246 (IntId256 VR256:$src1,
6247 (bitconvert (memopv32i8 addr:$src2))))]>, OpSize;
6250 let Predicates = [HasAVX] in {
6251 let isCommutable = 0 in
6252 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6254 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6256 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6258 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6260 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6262 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6264 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6266 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6268 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6270 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6272 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6275 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6276 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6277 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6278 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6281 let Predicates = [HasAVX2] in {
6282 let isCommutable = 0 in
6283 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6284 int_x86_avx2_packusdw>, VEX_4V;
6285 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6286 int_x86_avx2_pcmpeq_q>, VEX_4V;
6287 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6288 int_x86_avx2_pmins_b>, VEX_4V;
6289 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6290 int_x86_avx2_pmins_d>, VEX_4V;
6291 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6292 int_x86_avx2_pminu_d>, VEX_4V;
6293 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6294 int_x86_avx2_pminu_w>, VEX_4V;
6295 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6296 int_x86_avx2_pmaxs_b>, VEX_4V;
6297 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6298 int_x86_avx2_pmaxs_d>, VEX_4V;
6299 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6300 int_x86_avx2_pmaxu_d>, VEX_4V;
6301 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6302 int_x86_avx2_pmaxu_w>, VEX_4V;
6303 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6304 int_x86_avx2_pmul_dq>, VEX_4V;
6306 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6307 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6308 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6309 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6312 let Constraints = "$src1 = $dst" in {
6313 let isCommutable = 0 in
6314 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6315 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6316 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6317 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6318 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6319 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6320 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6321 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6322 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6323 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6324 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6327 let Predicates = [HasSSE41] in {
6328 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6329 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6330 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6331 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6334 /// SS48I_binop_rm - Simple SSE41 binary operator.
6335 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6336 ValueType OpVT, bit Is2Addr = 1> {
6337 let isCommutable = 1 in
6338 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6339 (ins VR128:$src1, VR128:$src2),
6341 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6342 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6343 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6345 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6346 (ins VR128:$src1, i128mem:$src2),
6348 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6349 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6350 [(set VR128:$dst, (OpNode VR128:$src1,
6351 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6355 /// SS48I_binop_rm - Simple SSE41 binary operator.
6356 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6358 let isCommutable = 1 in
6359 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6360 (ins VR256:$src1, VR256:$src2),
6361 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6362 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6364 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6365 (ins VR256:$src1, i256mem:$src2),
6366 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6367 [(set VR256:$dst, (OpNode VR256:$src1,
6368 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6372 let Predicates = [HasAVX] in
6373 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6374 let Predicates = [HasAVX2] in
6375 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6376 let Constraints = "$src1 = $dst" in
6377 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6379 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6380 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6381 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6382 X86MemOperand x86memop, bit Is2Addr = 1> {
6383 let isCommutable = 1 in
6384 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6385 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6387 !strconcat(OpcodeStr,
6388 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6389 !strconcat(OpcodeStr,
6390 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6391 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6393 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6394 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6396 !strconcat(OpcodeStr,
6397 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6398 !strconcat(OpcodeStr,
6399 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6402 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6406 let Predicates = [HasAVX] in {
6407 let isCommutable = 0 in {
6408 let ExeDomain = SSEPackedSingle in {
6409 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6410 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6411 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6412 int_x86_avx_blend_ps_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
6414 let ExeDomain = SSEPackedDouble in {
6415 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6416 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6417 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6418 int_x86_avx_blend_pd_256, VR256, memopv32i8, i256mem, 0>, VEX_4V;
6420 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6421 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6422 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6423 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6425 let ExeDomain = SSEPackedSingle in
6426 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6427 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6428 let ExeDomain = SSEPackedDouble in
6429 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6430 VR128, memopv16i8, i128mem, 0>, VEX_4V;
6431 let ExeDomain = SSEPackedSingle in
6432 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6433 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6436 let Predicates = [HasAVX2] in {
6437 let isCommutable = 0 in {
6438 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6439 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6440 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6441 VR256, memopv32i8, i256mem, 0>, VEX_4V;
6445 let Constraints = "$src1 = $dst" in {
6446 let isCommutable = 0 in {
6447 let ExeDomain = SSEPackedSingle in
6448 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6449 VR128, memopv16i8, i128mem>;
6450 let ExeDomain = SSEPackedDouble in
6451 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6452 VR128, memopv16i8, i128mem>;
6453 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6454 VR128, memopv16i8, i128mem>;
6455 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6456 VR128, memopv16i8, i128mem>;
6458 let ExeDomain = SSEPackedSingle in
6459 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6460 VR128, memopv16i8, i128mem>;
6461 let ExeDomain = SSEPackedDouble in
6462 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6463 VR128, memopv16i8, i128mem>;
6466 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6467 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6468 RegisterClass RC, X86MemOperand x86memop,
6469 PatFrag mem_frag, Intrinsic IntId> {
6470 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
6471 (ins RC:$src1, RC:$src2, RC:$src3),
6472 !strconcat(OpcodeStr,
6473 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6474 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6475 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6477 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
6478 (ins RC:$src1, x86memop:$src2, RC:$src3),
6479 !strconcat(OpcodeStr,
6480 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6482 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6484 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6487 let Predicates = [HasAVX] in {
6488 let ExeDomain = SSEPackedDouble in {
6489 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6490 memopv16i8, int_x86_sse41_blendvpd>;
6491 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6492 memopv32i8, int_x86_avx_blendv_pd_256>;
6493 } // ExeDomain = SSEPackedDouble
6494 let ExeDomain = SSEPackedSingle in {
6495 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6496 memopv16i8, int_x86_sse41_blendvps>;
6497 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6498 memopv32i8, int_x86_avx_blendv_ps_256>;
6499 } // ExeDomain = SSEPackedSingle
6500 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6501 memopv16i8, int_x86_sse41_pblendvb>;
6504 let Predicates = [HasAVX2] in {
6505 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6506 memopv32i8, int_x86_avx2_pblendvb>;
6509 let Predicates = [HasAVX] in {
6510 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6511 (v16i8 VR128:$src2))),
6512 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6513 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6514 (v4i32 VR128:$src2))),
6515 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6516 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6517 (v4f32 VR128:$src2))),
6518 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6519 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6520 (v2i64 VR128:$src2))),
6521 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6522 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6523 (v2f64 VR128:$src2))),
6524 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6525 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6526 (v8i32 VR256:$src2))),
6527 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6528 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6529 (v8f32 VR256:$src2))),
6530 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6531 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6532 (v4i64 VR256:$src2))),
6533 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6534 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6535 (v4f64 VR256:$src2))),
6536 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6539 let Predicates = [HasAVX2] in {
6540 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6541 (v32i8 VR256:$src2))),
6542 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6545 /// SS41I_ternary_int - SSE 4.1 ternary operator
6546 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6547 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
6548 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6549 (ins VR128:$src1, VR128:$src2),
6550 !strconcat(OpcodeStr,
6551 "\t{$src2, $dst|$dst, $src2}"),
6552 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6555 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6556 (ins VR128:$src1, i128mem:$src2),
6557 !strconcat(OpcodeStr,
6558 "\t{$src2, $dst|$dst, $src2}"),
6561 (bitconvert (memopv16i8 addr:$src2)), XMM0))]>, OpSize;
6565 let ExeDomain = SSEPackedDouble in
6566 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", int_x86_sse41_blendvpd>;
6567 let ExeDomain = SSEPackedSingle in
6568 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", int_x86_sse41_blendvps>;
6569 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", int_x86_sse41_pblendvb>;
6571 let Predicates = [HasSSE41] in {
6572 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6573 (v16i8 VR128:$src2))),
6574 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6575 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6576 (v4i32 VR128:$src2))),
6577 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6578 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6579 (v4f32 VR128:$src2))),
6580 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6581 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6582 (v2i64 VR128:$src2))),
6583 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6584 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6585 (v2f64 VR128:$src2))),
6586 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6589 let Predicates = [HasAVX] in
6590 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6591 "vmovntdqa\t{$src, $dst|$dst, $src}",
6592 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6594 let Predicates = [HasAVX2] in
6595 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6596 "vmovntdqa\t{$src, $dst|$dst, $src}",
6597 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6599 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6600 "movntdqa\t{$src, $dst|$dst, $src}",
6601 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6604 //===----------------------------------------------------------------------===//
6605 // SSE4.2 - Compare Instructions
6606 //===----------------------------------------------------------------------===//
6608 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6609 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6610 Intrinsic IntId128, bit Is2Addr = 1> {
6611 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6612 (ins VR128:$src1, VR128:$src2),
6614 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6615 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6616 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6618 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6619 (ins VR128:$src1, i128mem:$src2),
6621 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6622 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6624 (IntId128 VR128:$src1,
6625 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6628 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6629 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6630 Intrinsic IntId256> {
6631 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6632 (ins VR256:$src1, VR256:$src2),
6633 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6634 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6636 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6637 (ins VR256:$src1, i256mem:$src2),
6638 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6640 (IntId256 VR256:$src1,
6641 (bitconvert (memopv32i8 addr:$src2))))]>, OpSize;
6644 let Predicates = [HasAVX] in {
6645 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6648 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6649 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6650 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6651 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6654 let Predicates = [HasAVX2] in {
6655 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6658 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6659 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6660 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6661 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6664 let Constraints = "$src1 = $dst" in
6665 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6667 let Predicates = [HasSSE42] in {
6668 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6669 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6670 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6671 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6674 //===----------------------------------------------------------------------===//
6675 // SSE4.2 - String/text Processing Instructions
6676 //===----------------------------------------------------------------------===//
6678 // Packed Compare Implicit Length Strings, Return Mask
6679 multiclass pseudo_pcmpistrm<string asm> {
6680 def REG : PseudoI<(outs VR128:$dst),
6681 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6682 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6684 def MEM : PseudoI<(outs VR128:$dst),
6685 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6686 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6687 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6690 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6691 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6692 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6695 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6696 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6697 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6698 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6700 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6701 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6702 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6705 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6706 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6707 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6708 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6710 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6711 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6712 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6715 // Packed Compare Explicit Length Strings, Return Mask
6716 multiclass pseudo_pcmpestrm<string asm> {
6717 def REG : PseudoI<(outs VR128:$dst),
6718 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6719 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6720 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6721 def MEM : PseudoI<(outs VR128:$dst),
6722 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6723 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6724 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6727 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6728 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6729 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6732 let Predicates = [HasAVX],
6733 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6734 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6735 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6736 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6738 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6739 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6740 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6743 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6744 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6745 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6746 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6748 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6749 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6750 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6753 // Packed Compare Implicit Length Strings, Return Index
6754 let Defs = [ECX, EFLAGS] in {
6755 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6756 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6757 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6758 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6759 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6760 (implicit EFLAGS)]>, OpSize;
6761 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6762 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6763 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6764 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6765 (implicit EFLAGS)]>, OpSize;
6769 let Predicates = [HasAVX] in {
6770 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6772 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6774 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6776 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6778 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6780 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6784 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6785 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6786 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6787 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6788 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6789 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6791 // Packed Compare Explicit Length Strings, Return Index
6792 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6793 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6794 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6795 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6796 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6797 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6798 (implicit EFLAGS)]>, OpSize;
6799 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6800 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6801 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6803 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6804 (implicit EFLAGS)]>, OpSize;
6808 let Predicates = [HasAVX] in {
6809 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6811 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6813 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6815 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6817 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6819 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6823 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6824 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6825 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6826 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6827 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6828 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6830 //===----------------------------------------------------------------------===//
6831 // SSE4.2 - CRC Instructions
6832 //===----------------------------------------------------------------------===//
6834 // No CRC instructions have AVX equivalents
6836 // crc intrinsic instruction
6837 // This set of instructions are only rm, the only difference is the size
6839 let Constraints = "$src1 = $dst" in {
6840 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6841 (ins GR32:$src1, i8mem:$src2),
6842 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6844 (int_x86_sse42_crc32_32_8 GR32:$src1,
6845 (load addr:$src2)))]>;
6846 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6847 (ins GR32:$src1, GR8:$src2),
6848 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6850 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6851 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6852 (ins GR32:$src1, i16mem:$src2),
6853 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6855 (int_x86_sse42_crc32_32_16 GR32:$src1,
6856 (load addr:$src2)))]>,
6858 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6859 (ins GR32:$src1, GR16:$src2),
6860 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6862 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6864 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6865 (ins GR32:$src1, i32mem:$src2),
6866 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6868 (int_x86_sse42_crc32_32_32 GR32:$src1,
6869 (load addr:$src2)))]>;
6870 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6871 (ins GR32:$src1, GR32:$src2),
6872 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6874 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6875 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6876 (ins GR64:$src1, i8mem:$src2),
6877 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6879 (int_x86_sse42_crc32_64_8 GR64:$src1,
6880 (load addr:$src2)))]>,
6882 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6883 (ins GR64:$src1, GR8:$src2),
6884 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6886 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6888 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6889 (ins GR64:$src1, i64mem:$src2),
6890 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6892 (int_x86_sse42_crc32_64_64 GR64:$src1,
6893 (load addr:$src2)))]>,
6895 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6896 (ins GR64:$src1, GR64:$src2),
6897 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6899 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
6903 //===----------------------------------------------------------------------===//
6904 // AES-NI Instructions
6905 //===----------------------------------------------------------------------===//
6907 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
6908 Intrinsic IntId128, bit Is2Addr = 1> {
6909 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
6910 (ins VR128:$src1, VR128:$src2),
6912 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6913 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6914 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6916 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
6917 (ins VR128:$src1, i128mem:$src2),
6919 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6920 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6922 (IntId128 VR128:$src1,
6923 (bitconvert (memopv16i8 addr:$src2))))]>, OpSize;
6926 // Perform One Round of an AES Encryption/Decryption Flow
6927 let Predicates = [HasAVX, HasAES] in {
6928 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
6929 int_x86_aesni_aesenc, 0>, VEX_4V;
6930 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
6931 int_x86_aesni_aesenclast, 0>, VEX_4V;
6932 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
6933 int_x86_aesni_aesdec, 0>, VEX_4V;
6934 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
6935 int_x86_aesni_aesdeclast, 0>, VEX_4V;
6938 let Constraints = "$src1 = $dst" in {
6939 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
6940 int_x86_aesni_aesenc>;
6941 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
6942 int_x86_aesni_aesenclast>;
6943 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
6944 int_x86_aesni_aesdec>;
6945 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
6946 int_x86_aesni_aesdeclast>;
6949 let Predicates = [HasAES] in {
6950 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6951 (AESENCrr VR128:$src1, VR128:$src2)>;
6952 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6953 (AESENCrm VR128:$src1, addr:$src2)>;
6954 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
6955 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
6956 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
6957 (AESENCLASTrm VR128:$src1, addr:$src2)>;
6958 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
6959 (AESDECrr VR128:$src1, VR128:$src2)>;
6960 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
6961 (AESDECrm VR128:$src1, addr:$src2)>;
6962 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
6963 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
6964 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
6965 (AESDECLASTrm VR128:$src1, addr:$src2)>;
6968 let Predicates = [HasAVX, HasAES], AddedComplexity = 20 in {
6969 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
6970 (VAESENCrr VR128:$src1, VR128:$src2)>;
6971 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
6972 (VAESENCrm VR128:$src1, addr:$src2)>;
6973 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
6974 (VAESENCLASTrr VR128:$src1, VR128:$src2)>;
6975 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
6976 (VAESENCLASTrm VR128:$src1, addr:$src2)>;
6977 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
6978 (VAESDECrr VR128:$src1, VR128:$src2)>;
6979 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
6980 (VAESDECrm VR128:$src1, addr:$src2)>;
6981 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
6982 (VAESDECLASTrr VR128:$src1, VR128:$src2)>;
6983 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
6984 (VAESDECLASTrm VR128:$src1, addr:$src2)>;
6987 // Perform the AES InvMixColumn Transformation
6988 let Predicates = [HasAVX, HasAES] in {
6989 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
6991 "vaesimc\t{$src1, $dst|$dst, $src1}",
6993 (int_x86_aesni_aesimc VR128:$src1))]>,
6995 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
6996 (ins i128mem:$src1),
6997 "vaesimc\t{$src1, $dst|$dst, $src1}",
6999 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7002 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7004 "aesimc\t{$src1, $dst|$dst, $src1}",
7006 (int_x86_aesni_aesimc VR128:$src1))]>,
7008 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7009 (ins i128mem:$src1),
7010 "aesimc\t{$src1, $dst|$dst, $src1}",
7012 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7015 // AES Round Key Generation Assist
7016 let Predicates = [HasAVX, HasAES] in {
7017 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7018 (ins VR128:$src1, i8imm:$src2),
7019 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7021 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7023 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7024 (ins i128mem:$src1, i8imm:$src2),
7025 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7027 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7031 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7032 (ins VR128:$src1, i8imm:$src2),
7033 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7035 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7037 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7038 (ins i128mem:$src1, i8imm:$src2),
7039 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7041 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7045 //===----------------------------------------------------------------------===//
7046 // CLMUL Instructions
7047 //===----------------------------------------------------------------------===//
7049 // Carry-less Multiplication instructions
7050 let neverHasSideEffects = 1 in {
7051 let Constraints = "$src1 = $dst" in {
7052 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7053 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7054 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7058 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7059 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7060 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7064 // AVX carry-less Multiplication instructions
7065 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7066 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7067 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7071 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7072 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7073 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7078 multiclass pclmul_alias<string asm, int immop> {
7079 def : InstAlias<!strconcat("pclmul", asm,
7080 "dq {$src, $dst|$dst, $src}"),
7081 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7083 def : InstAlias<!strconcat("pclmul", asm,
7084 "dq {$src, $dst|$dst, $src}"),
7085 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7087 def : InstAlias<!strconcat("vpclmul", asm,
7088 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7089 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7091 def : InstAlias<!strconcat("vpclmul", asm,
7092 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7093 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7095 defm : pclmul_alias<"hqhq", 0x11>;
7096 defm : pclmul_alias<"hqlq", 0x01>;
7097 defm : pclmul_alias<"lqhq", 0x10>;
7098 defm : pclmul_alias<"lqlq", 0x00>;
7100 //===----------------------------------------------------------------------===//
7102 //===----------------------------------------------------------------------===//
7104 //===----------------------------------------------------------------------===//
7105 // VBROADCAST - Load from memory and broadcast to all elements of the
7106 // destination operand
7108 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7109 X86MemOperand x86memop, Intrinsic Int> :
7110 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7111 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7112 [(set RC:$dst, (Int addr:$src))]>, VEX;
7114 // AVX2 adds register forms
7115 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7117 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7118 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7119 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7121 let ExeDomain = SSEPackedSingle in {
7122 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7123 int_x86_avx_vbroadcast_ss>;
7124 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7125 int_x86_avx_vbroadcast_ss_256>;
7127 let ExeDomain = SSEPackedDouble in
7128 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7129 int_x86_avx_vbroadcast_sd_256>;
7130 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7131 int_x86_avx_vbroadcastf128_pd_256>;
7133 let ExeDomain = SSEPackedSingle in {
7134 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7135 int_x86_avx2_vbroadcast_ss_ps>;
7136 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7137 int_x86_avx2_vbroadcast_ss_ps_256>;
7139 let ExeDomain = SSEPackedDouble in
7140 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7141 int_x86_avx2_vbroadcast_sd_pd_256>;
7143 let Predicates = [HasAVX2] in
7144 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7145 int_x86_avx2_vbroadcasti128>;
7147 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7148 (VBROADCASTF128 addr:$src)>;
7151 //===----------------------------------------------------------------------===//
7152 // VINSERTF128 - Insert packed floating-point values
7154 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7155 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7156 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7157 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7160 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7161 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7162 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7166 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7167 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7168 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7169 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7170 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7171 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7173 //===----------------------------------------------------------------------===//
7174 // VEXTRACTF128 - Extract packed floating-point values
7176 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7177 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7178 (ins VR256:$src1, i8imm:$src2),
7179 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7182 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7183 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7184 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7188 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7189 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7190 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7191 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7192 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7193 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7195 //===----------------------------------------------------------------------===//
7196 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7198 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7199 Intrinsic IntLd, Intrinsic IntLd256,
7200 Intrinsic IntSt, Intrinsic IntSt256> {
7201 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7202 (ins VR128:$src1, f128mem:$src2),
7203 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7204 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7206 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7207 (ins VR256:$src1, f256mem:$src2),
7208 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7209 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7211 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7212 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7213 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7214 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7215 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7216 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7218 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7221 let ExeDomain = SSEPackedSingle in
7222 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7223 int_x86_avx_maskload_ps,
7224 int_x86_avx_maskload_ps_256,
7225 int_x86_avx_maskstore_ps,
7226 int_x86_avx_maskstore_ps_256>;
7227 let ExeDomain = SSEPackedDouble in
7228 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7229 int_x86_avx_maskload_pd,
7230 int_x86_avx_maskload_pd_256,
7231 int_x86_avx_maskstore_pd,
7232 int_x86_avx_maskstore_pd_256>;
7234 //===----------------------------------------------------------------------===//
7235 // VPERMIL - Permute Single and Double Floating-Point Values
7237 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7238 RegisterClass RC, X86MemOperand x86memop_f,
7239 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7240 Intrinsic IntVar, Intrinsic IntImm> {
7241 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7242 (ins RC:$src1, RC:$src2),
7243 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7244 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7245 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7246 (ins RC:$src1, x86memop_i:$src2),
7247 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7248 [(set RC:$dst, (IntVar RC:$src1, (i_frag addr:$src2)))]>, VEX_4V;
7250 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7251 (ins RC:$src1, i8imm:$src2),
7252 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7253 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7254 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7255 (ins x86memop_f:$src1, i8imm:$src2),
7256 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7257 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7260 let ExeDomain = SSEPackedSingle in {
7261 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7262 memopv4f32, memopv4i32,
7263 int_x86_avx_vpermilvar_ps,
7264 int_x86_avx_vpermil_ps>;
7265 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7266 memopv8f32, memopv8i32,
7267 int_x86_avx_vpermilvar_ps_256,
7268 int_x86_avx_vpermil_ps_256>;
7270 let ExeDomain = SSEPackedDouble in {
7271 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7272 memopv2f64, memopv2i64,
7273 int_x86_avx_vpermilvar_pd,
7274 int_x86_avx_vpermil_pd>;
7275 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7276 memopv4f64, memopv4i64,
7277 int_x86_avx_vpermilvar_pd_256,
7278 int_x86_avx_vpermil_pd_256>;
7281 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7282 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7283 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7284 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7285 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7286 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7287 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7288 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7289 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7290 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7291 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7292 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7293 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7295 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7296 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7297 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7299 //===----------------------------------------------------------------------===//
7300 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7302 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7303 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7304 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7305 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7308 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7309 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7310 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7314 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7315 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7316 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7317 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7318 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7319 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7321 def : Pat<(int_x86_avx_vperm2f128_ps_256
7322 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7323 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7324 def : Pat<(int_x86_avx_vperm2f128_pd_256
7325 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7326 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7327 def : Pat<(int_x86_avx_vperm2f128_si_256
7328 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7329 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7331 //===----------------------------------------------------------------------===//
7332 // VZERO - Zero YMM registers
7334 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7335 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7336 // Zero All YMM registers
7337 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7338 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7340 // Zero Upper bits of YMM registers
7341 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7342 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7345 //===----------------------------------------------------------------------===//
7346 // Half precision conversion instructions
7347 //===----------------------------------------------------------------------===//
7348 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7349 let Predicates = [HasAVX, HasF16C] in {
7350 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7351 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7352 [(set RC:$dst, (Int VR128:$src))]>,
7354 let neverHasSideEffects = 1, mayLoad = 1 in
7355 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7356 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7360 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7361 let Predicates = [HasAVX, HasF16C] in {
7362 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7363 (ins RC:$src1, i32i8imm:$src2),
7364 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7365 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7367 let neverHasSideEffects = 1, mayLoad = 1 in
7368 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7369 (ins RC:$src1, i32i8imm:$src2),
7370 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7375 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7376 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7377 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7378 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7380 //===----------------------------------------------------------------------===//
7381 // AVX2 Instructions
7382 //===----------------------------------------------------------------------===//
7384 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7385 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7386 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7387 X86MemOperand x86memop> {
7388 let isCommutable = 1 in
7389 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7390 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7391 !strconcat(OpcodeStr,
7392 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7393 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7395 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7396 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7397 !strconcat(OpcodeStr,
7398 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7401 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7405 let isCommutable = 0 in {
7406 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7407 VR128, memopv16i8, i128mem>;
7408 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7409 VR256, memopv32i8, i256mem>;
7412 //===----------------------------------------------------------------------===//
7413 // VPBROADCAST - Load from memory and broadcast to all elements of the
7414 // destination operand
7416 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7417 X86MemOperand x86memop, PatFrag ld_frag,
7418 Intrinsic Int128, Intrinsic Int256> {
7419 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7420 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7421 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7422 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7423 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7425 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7426 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7427 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7428 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7429 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7430 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7432 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7435 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7436 int_x86_avx2_pbroadcastb_128,
7437 int_x86_avx2_pbroadcastb_256>;
7438 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7439 int_x86_avx2_pbroadcastw_128,
7440 int_x86_avx2_pbroadcastw_256>;
7441 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7442 int_x86_avx2_pbroadcastd_128,
7443 int_x86_avx2_pbroadcastd_256>;
7444 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7445 int_x86_avx2_pbroadcastq_128,
7446 int_x86_avx2_pbroadcastq_256>;
7448 let Predicates = [HasAVX2] in {
7449 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7450 (VPBROADCASTBrm addr:$src)>;
7451 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7452 (VPBROADCASTBYrm addr:$src)>;
7453 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7454 (VPBROADCASTWrm addr:$src)>;
7455 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7456 (VPBROADCASTWYrm addr:$src)>;
7457 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7458 (VPBROADCASTDrm addr:$src)>;
7459 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7460 (VPBROADCASTDYrm addr:$src)>;
7461 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7462 (VPBROADCASTQrm addr:$src)>;
7463 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7464 (VPBROADCASTQYrm addr:$src)>;
7467 // AVX1 broadcast patterns
7468 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7469 (VBROADCASTSSYrm addr:$src)>;
7470 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7471 (VBROADCASTSDrm addr:$src)>;
7472 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7473 (VBROADCASTSSYrm addr:$src)>;
7474 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7475 (VBROADCASTSDrm addr:$src)>;
7477 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7478 (VBROADCASTSSrm addr:$src)>;
7479 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7480 (VBROADCASTSSrm addr:$src)>;
7482 //===----------------------------------------------------------------------===//
7483 // VPERM - Permute instructions
7486 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7488 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7489 (ins VR256:$src1, VR256:$src2),
7490 !strconcat(OpcodeStr,
7491 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7492 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7493 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7494 (ins VR256:$src1, i256mem:$src2),
7495 !strconcat(OpcodeStr,
7496 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7497 [(set VR256:$dst, (Int VR256:$src1, (mem_frag addr:$src2)))]>,
7501 defm VPERMD : avx2_perm<0x36, "vpermd", memopv8i32, int_x86_avx2_permd>;
7502 let ExeDomain = SSEPackedSingle in
7503 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7505 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7507 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7508 (ins VR256:$src1, i8imm:$src2),
7509 !strconcat(OpcodeStr,
7510 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7511 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7512 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7513 (ins i256mem:$src1, i8imm:$src2),
7514 !strconcat(OpcodeStr,
7515 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7516 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7520 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7522 let ExeDomain = SSEPackedDouble in
7523 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7526 //===----------------------------------------------------------------------===//
7527 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7529 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7530 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7531 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7533 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7535 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7536 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7537 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7539 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7543 let Predicates = [HasAVX2] in {
7544 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7545 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7546 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7547 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7548 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7549 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7550 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7551 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7553 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7555 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7556 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7557 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7558 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7559 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7561 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7562 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7564 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7568 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7569 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7570 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7571 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7572 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7573 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7574 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7575 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7576 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7577 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7578 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7579 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7581 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7582 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7583 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7584 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7585 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7586 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7587 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7588 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7589 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7590 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7591 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7592 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7593 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7594 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7595 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7596 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7597 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7598 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7601 //===----------------------------------------------------------------------===//
7602 // VINSERTI128 - Insert packed integer values
7604 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7605 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7606 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7608 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7610 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7611 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7612 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7614 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7615 imm:$src3))]>, VEX_4V;
7617 let Predicates = [HasAVX2] in {
7618 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7620 (VINSERTI128rr VR256:$src1, VR128:$src2,
7621 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7622 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7624 (VINSERTI128rr VR256:$src1, VR128:$src2,
7625 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7626 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7628 (VINSERTI128rr VR256:$src1, VR128:$src2,
7629 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7630 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7632 (VINSERTI128rr VR256:$src1, VR128:$src2,
7633 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7637 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7639 (VINSERTF128rr VR256:$src1, VR128:$src2,
7640 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7641 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7643 (VINSERTF128rr VR256:$src1, VR128:$src2,
7644 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7645 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7647 (VINSERTF128rr VR256:$src1, VR128:$src2,
7648 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7649 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7651 (VINSERTF128rr VR256:$src1, VR128:$src2,
7652 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7653 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7655 (VINSERTF128rr VR256:$src1, VR128:$src2,
7656 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7657 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7659 (VINSERTF128rr VR256:$src1, VR128:$src2,
7660 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7662 //===----------------------------------------------------------------------===//
7663 // VEXTRACTI128 - Extract packed integer values
7665 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7666 (ins VR256:$src1, i8imm:$src2),
7667 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7669 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7671 let neverHasSideEffects = 1, mayStore = 1 in
7672 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7673 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7674 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7676 let Predicates = [HasAVX2] in {
7677 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7678 (v2i64 (VEXTRACTI128rr
7679 (v4i64 VR256:$src1),
7680 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7681 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7682 (v4i32 (VEXTRACTI128rr
7683 (v8i32 VR256:$src1),
7684 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7685 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7686 (v8i16 (VEXTRACTI128rr
7687 (v16i16 VR256:$src1),
7688 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7689 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7690 (v16i8 (VEXTRACTI128rr
7691 (v32i8 VR256:$src1),
7692 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7696 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7697 (v4f32 (VEXTRACTF128rr
7698 (v8f32 VR256:$src1),
7699 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7700 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7701 (v2f64 (VEXTRACTF128rr
7702 (v4f64 VR256:$src1),
7703 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7704 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7705 (v2i64 (VEXTRACTF128rr
7706 (v4i64 VR256:$src1),
7707 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7708 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7709 (v4i32 (VEXTRACTF128rr
7710 (v8i32 VR256:$src1),
7711 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7712 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7713 (v8i16 (VEXTRACTF128rr
7714 (v16i16 VR256:$src1),
7715 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7716 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7717 (v16i8 (VEXTRACTF128rr
7718 (v32i8 VR256:$src1),
7719 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7721 //===----------------------------------------------------------------------===//
7722 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7724 multiclass avx2_pmovmask<string OpcodeStr,
7725 Intrinsic IntLd128, Intrinsic IntLd256,
7726 Intrinsic IntSt128, Intrinsic IntSt256> {
7727 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7728 (ins VR128:$src1, i128mem:$src2),
7729 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7730 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7731 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7732 (ins VR256:$src1, i256mem:$src2),
7733 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7734 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7735 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7736 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7737 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7738 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7739 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7740 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7741 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7742 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7745 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7746 int_x86_avx2_maskload_d,
7747 int_x86_avx2_maskload_d_256,
7748 int_x86_avx2_maskstore_d,
7749 int_x86_avx2_maskstore_d_256>;
7750 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7751 int_x86_avx2_maskload_q,
7752 int_x86_avx2_maskload_q_256,
7753 int_x86_avx2_maskstore_q,
7754 int_x86_avx2_maskstore_q_256>, VEX_W;
7757 //===----------------------------------------------------------------------===//
7758 // Variable Bit Shifts
7760 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7761 ValueType vt128, ValueType vt256> {
7762 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7763 (ins VR128:$src1, VR128:$src2),
7764 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7766 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7768 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7769 (ins VR128:$src1, i128mem:$src2),
7770 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7772 (vt128 (OpNode VR128:$src1,
7773 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7775 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7776 (ins VR256:$src1, VR256:$src2),
7777 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7779 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7781 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7782 (ins VR256:$src1, i256mem:$src2),
7783 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7785 (vt256 (OpNode VR256:$src1,
7786 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7790 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7791 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7792 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7793 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7794 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;