1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 bit rr_hasSideEffects = 0> {
85 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
86 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
88 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
89 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
93 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
94 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
98 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
99 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
100 string asm, string SSEVer, string FPSizeStr,
101 X86MemOperand x86memop, PatFrag mem_frag,
102 Domain d, bit Is2Addr = 1> {
103 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
105 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
106 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
107 [(set RC:$dst, (!cast<Intrinsic>(
108 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
109 RC:$src1, RC:$src2))], d>;
110 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
112 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
113 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
114 [(set RC:$dst, (!cast<Intrinsic>(
115 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
116 RC:$src1, (mem_frag addr:$src2)))], d>;
119 //===----------------------------------------------------------------------===//
120 // Non-instruction patterns
121 //===----------------------------------------------------------------------===//
123 // A vector extract of the first f32/f64 position is a subregister copy
124 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
125 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
127 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
129 // A 128-bit subvector extract from the first 256-bit vector position
130 // is a subregister copy that needs no instruction.
131 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
132 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
133 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
134 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
136 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
137 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
138 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
139 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
141 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
142 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
143 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
144 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
146 // A 128-bit subvector insert to the first 256-bit vector position
147 // is a subregister copy that needs no instruction.
148 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
149 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
150 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
151 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
152 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
153 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
154 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
155 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
156 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
157 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
159 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
161 // Implicitly promote a 32-bit scalar to a vector.
162 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
163 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
164 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
165 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
166 // Implicitly promote a 64-bit scalar to a vector.
167 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
168 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
170 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
172 // Bitcasts between 128-bit vector types. Return the original type since
173 // no instruction is needed for the conversion
174 let Predicates = [HasXMMInt] in {
175 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
207 // Bitcasts between 256-bit vector types. Return the original type since
208 // no instruction is needed for the conversion
209 let Predicates = [HasAVX] in {
210 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
242 // Alias instructions that map fld0 to pxor for sse.
243 // This is expanded by ExpandPostRAPseudos.
244 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
246 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
247 [(set FR32:$dst, fp32imm0)]>, Requires<[HasXMM]>;
248 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
249 [(set FR64:$dst, fpimm0)]>, Requires<[HasXMMInt]>;
252 //===----------------------------------------------------------------------===//
253 // AVX & SSE - Zero/One Vectors
254 //===----------------------------------------------------------------------===//
256 // Alias instruction that maps zero vector to pxor / xorp* for sse.
257 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
258 // swizzled by ExecutionDepsFix to pxor.
259 // We set canFoldAsLoad because this can be converted to a constant-pool
260 // load of an all-zeros value if folding it would be beneficial.
261 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
262 isPseudo = 1, neverHasSideEffects = 1 in {
263 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
266 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
267 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
268 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
269 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
270 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
271 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
274 // The same as done above but for AVX. The 256-bit ISA does not support PI,
275 // and doesn't need it because on sandy bridge the register is set to zero
276 // at the rename stage without using any execution unit, so SET0PSY
277 // and SET0PDY can be used for vector int instructions without penalty
278 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
279 // JIT implementatioan, it does not expand the instructions below like
280 // X86MCInstLower does.
281 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
282 isCodeGenOnly = 1, Predicates = [HasAVX] in {
283 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
284 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
285 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 // AVX has no support for 256-bit integer instructions, but since the 128-bit
291 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
292 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
293 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
294 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
296 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
297 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
298 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
300 // We set canFoldAsLoad because this can be converted to a constant-pool
301 // load of an all-ones value if folding it would be beneficial.
302 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
303 // JIT implementation, it does not expand the instructions below like
304 // X86MCInstLower does.
305 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
306 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
307 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
308 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
309 let Predicates = [HasAVX] in
310 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
311 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
312 let Predicates = [HasAVX2] in
313 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
314 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
318 //===----------------------------------------------------------------------===//
319 // SSE 1 & 2 - Move FP Scalar Instructions
321 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
322 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
323 // is used instead. Register-to-register movss/movsd is not modeled as an
324 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
325 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
326 //===----------------------------------------------------------------------===//
328 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
329 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
330 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
332 // Loading from memory automatically zeroing upper bits.
333 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
334 PatFrag mem_pat, string OpcodeStr> :
335 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
337 [(set RC:$dst, (mem_pat addr:$src))]>;
340 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
341 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
343 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
344 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
347 // For the disassembler
348 let isCodeGenOnly = 1 in {
349 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
350 (ins VR128:$src1, FR32:$src2),
351 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
353 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
354 (ins VR128:$src1, FR64:$src2),
355 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
359 let canFoldAsLoad = 1, isReMaterializable = 1 in {
360 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
362 let AddedComplexity = 20 in
363 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
367 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
368 "movss\t{$src, $dst|$dst, $src}",
369 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
370 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
371 "movsd\t{$src, $dst|$dst, $src}",
372 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
375 let Constraints = "$src1 = $dst" in {
376 def MOVSSrr : sse12_move_rr<FR32, v4f32,
377 "movss\t{$src2, $dst|$dst, $src2}">, XS;
378 def MOVSDrr : sse12_move_rr<FR64, v2f64,
379 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
381 // For the disassembler
382 let isCodeGenOnly = 1 in {
383 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
384 (ins VR128:$src1, FR32:$src2),
385 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
386 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
387 (ins VR128:$src1, FR64:$src2),
388 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
392 let canFoldAsLoad = 1, isReMaterializable = 1 in {
393 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
395 let AddedComplexity = 20 in
396 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
399 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
400 "movss\t{$src, $dst|$dst, $src}",
401 [(store FR32:$src, addr:$dst)]>;
402 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
403 "movsd\t{$src, $dst|$dst, $src}",
404 [(store FR64:$src, addr:$dst)]>;
407 let Predicates = [HasSSE1] in {
408 let AddedComplexity = 15 in {
409 // Extract the low 32-bit value from one vector and insert it into another.
410 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
411 (MOVSSrr (v4f32 VR128:$src1),
412 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
413 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
414 (MOVSSrr (v4i32 VR128:$src1),
415 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
417 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
418 // MOVSS to the lower bits.
419 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
420 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
421 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
422 (MOVSSrr (v4f32 (V_SET0)),
423 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
424 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
425 (MOVSSrr (v4i32 (V_SET0)),
426 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
429 let AddedComplexity = 20 in {
430 // MOVSSrm zeros the high parts of the register; represent this
431 // with SUBREG_TO_REG.
432 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
433 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
434 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
435 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
436 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
437 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
440 // Extract and store.
441 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
444 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
446 // Shuffle with MOVSS
447 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
448 (MOVSSrr VR128:$src1, FR32:$src2)>;
449 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
450 (MOVSSrr (v4i32 VR128:$src1),
451 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
452 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
453 (MOVSSrr (v4f32 VR128:$src1),
454 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
457 let Predicates = [HasSSE2] in {
458 let AddedComplexity = 15 in {
459 // Extract the low 64-bit value from one vector and insert it into another.
460 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
461 (MOVSDrr (v2f64 VR128:$src1),
462 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
463 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
464 (MOVSDrr (v2i64 VR128:$src1),
465 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
467 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
468 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
469 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
470 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
471 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
473 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
474 // MOVSD to the lower bits.
475 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
476 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
479 let AddedComplexity = 20 in {
480 // MOVSDrm zeros the high parts of the register; represent this
481 // with SUBREG_TO_REG.
482 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
483 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
484 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
485 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
486 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
487 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
488 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
489 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
490 def : Pat<(v2f64 (X86vzload addr:$src)),
491 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
494 // Extract and store.
495 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
498 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
500 // Shuffle with MOVSD
501 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
502 (MOVSDrr VR128:$src1, FR64:$src2)>;
503 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
504 (MOVSDrr (v2i64 VR128:$src1),
505 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
506 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
507 (MOVSDrr (v2f64 VR128:$src1),
508 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
509 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
510 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
511 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
512 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
514 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
515 // is during lowering, where it's not possible to recognize the fold cause
516 // it has two uses through a bitcast. One use disappears at isel time and the
517 // fold opportunity reappears.
518 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
519 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
520 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
521 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
522 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
523 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
524 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
525 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
528 let Predicates = [HasAVX] in {
529 let AddedComplexity = 15 in {
530 // Extract the low 32-bit value from one vector and insert it into another.
531 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
532 (VMOVSSrr (v4f32 VR128:$src1),
533 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
534 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
535 (VMOVSSrr (v4i32 VR128:$src1),
536 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
538 // Extract the low 64-bit value from one vector and insert it into another.
539 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
540 (VMOVSDrr (v2f64 VR128:$src1),
541 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
542 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
543 (VMOVSDrr (v2i64 VR128:$src1),
544 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
546 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
547 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
548 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
549 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
550 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
552 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
553 // MOVS{S,D} to the lower bits.
554 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
555 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
556 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
557 (VMOVSSrr (v4f32 (V_SET0)),
558 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
559 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
560 (VMOVSSrr (v4i32 (V_SET0)),
561 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
562 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
563 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
565 // Move low f32 and clear high bits.
566 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
567 (SUBREG_TO_REG (i32 0),
568 (VMOVSSrr (v4f32 (V_SET0)),
569 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
570 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
571 (SUBREG_TO_REG (i32 0),
572 (VMOVSSrr (v4i32 (V_SET0)),
573 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
576 let AddedComplexity = 20 in {
577 // MOVSSrm zeros the high parts of the register; represent this
578 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
579 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
581 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
582 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
583 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
584 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
586 // MOVSDrm zeros the high parts of the register; represent this
587 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
588 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
589 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
590 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
591 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
592 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
593 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
594 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
595 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
596 def : Pat<(v2f64 (X86vzload addr:$src)),
597 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
599 // Represent the same patterns above but in the form they appear for
601 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
602 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
603 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
604 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
605 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
606 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
607 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
608 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
609 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
611 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
612 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
613 (SUBREG_TO_REG (i32 0),
614 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
616 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
617 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
618 (SUBREG_TO_REG (i64 0),
619 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
621 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
622 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
623 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
625 // Move low f64 and clear high bits.
626 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
627 (SUBREG_TO_REG (i32 0),
628 (VMOVSDrr (v2f64 (V_SET0)),
629 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
631 // Extract and store.
632 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
635 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
636 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
639 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
641 // Shuffle with VMOVSS
642 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
643 (VMOVSSrr VR128:$src1, FR32:$src2)>;
644 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
645 (VMOVSSrr (v4i32 VR128:$src1),
646 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
647 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
648 (VMOVSSrr (v4f32 VR128:$src1),
649 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
652 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
653 (SUBREG_TO_REG (i32 0),
654 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
655 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
656 def : Pat<(v8f32 (X86Movsd VR256:$src1, VR256:$src2)),
657 (SUBREG_TO_REG (i32 0),
658 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
659 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
661 // Shuffle with VMOVSD
662 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
663 (VMOVSDrr VR128:$src1, FR64:$src2)>;
664 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
665 (VMOVSDrr (v2i64 VR128:$src1),
666 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
667 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
668 (VMOVSDrr (v2f64 VR128:$src1),
669 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
670 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
671 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
673 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
674 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
678 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
679 (SUBREG_TO_REG (i32 0),
680 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
681 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
682 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
683 (SUBREG_TO_REG (i32 0),
684 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
685 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
688 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
689 // is during lowering, where it's not possible to recognize the fold cause
690 // it has two uses through a bitcast. One use disappears at isel time and the
691 // fold opportunity reappears.
692 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
693 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
695 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
696 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
698 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
699 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
701 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
702 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
706 //===----------------------------------------------------------------------===//
707 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
708 //===----------------------------------------------------------------------===//
710 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
711 X86MemOperand x86memop, PatFrag ld_frag,
712 string asm, Domain d,
713 bit IsReMaterializable = 1> {
714 let neverHasSideEffects = 1 in
715 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
716 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
717 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
718 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
719 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
720 [(set RC:$dst, (ld_frag addr:$src))], d>;
723 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
724 "movaps", SSEPackedSingle>, TB, VEX;
725 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
726 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
727 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
728 "movups", SSEPackedSingle>, TB, VEX;
729 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
730 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
732 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
733 "movaps", SSEPackedSingle>, TB, VEX;
734 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
735 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
736 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
737 "movups", SSEPackedSingle>, TB, VEX;
738 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
739 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
740 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
741 "movaps", SSEPackedSingle>, TB;
742 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
743 "movapd", SSEPackedDouble>, TB, OpSize;
744 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
745 "movups", SSEPackedSingle>, TB;
746 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
747 "movupd", SSEPackedDouble, 0>, TB, OpSize;
749 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
750 "movaps\t{$src, $dst|$dst, $src}",
751 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
752 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
753 "movapd\t{$src, $dst|$dst, $src}",
754 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
755 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
756 "movups\t{$src, $dst|$dst, $src}",
757 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
758 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
759 "movupd\t{$src, $dst|$dst, $src}",
760 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
761 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
762 "movaps\t{$src, $dst|$dst, $src}",
763 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
764 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
765 "movapd\t{$src, $dst|$dst, $src}",
766 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
767 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
768 "movups\t{$src, $dst|$dst, $src}",
769 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
770 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
771 "movupd\t{$src, $dst|$dst, $src}",
772 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
775 let isCodeGenOnly = 1 in {
776 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
778 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
779 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
781 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
782 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
784 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
785 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
787 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
788 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
790 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
791 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
793 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
794 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
796 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
797 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
799 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
802 let Predicates = [HasAVX] in {
803 def : Pat<(v8i32 (X86vzmovl
804 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
805 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
806 def : Pat<(v4i64 (X86vzmovl
807 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
808 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
809 def : Pat<(v8f32 (X86vzmovl
810 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
811 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
812 def : Pat<(v4f64 (X86vzmovl
813 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
814 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
818 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
819 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
820 (VMOVUPSYmr addr:$dst, VR256:$src)>;
822 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
823 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
824 (VMOVUPDYmr addr:$dst, VR256:$src)>;
826 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
827 "movaps\t{$src, $dst|$dst, $src}",
828 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
829 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
830 "movapd\t{$src, $dst|$dst, $src}",
831 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
832 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
833 "movups\t{$src, $dst|$dst, $src}",
834 [(store (v4f32 VR128:$src), addr:$dst)]>;
835 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
836 "movupd\t{$src, $dst|$dst, $src}",
837 [(store (v2f64 VR128:$src), addr:$dst)]>;
840 let isCodeGenOnly = 1 in {
841 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
842 "movaps\t{$src, $dst|$dst, $src}", []>;
843 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
844 "movapd\t{$src, $dst|$dst, $src}", []>;
845 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
846 "movups\t{$src, $dst|$dst, $src}", []>;
847 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
848 "movupd\t{$src, $dst|$dst, $src}", []>;
851 let Predicates = [HasAVX] in {
852 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
853 (VMOVUPSmr addr:$dst, VR128:$src)>;
854 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
855 (VMOVUPDmr addr:$dst, VR128:$src)>;
858 let Predicates = [HasSSE1] in
859 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
860 (MOVUPSmr addr:$dst, VR128:$src)>;
861 let Predicates = [HasSSE2] in
862 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
863 (MOVUPDmr addr:$dst, VR128:$src)>;
865 // Use movaps / movups for SSE integer load / store (one byte shorter).
866 // The instructions selected below are then converted to MOVDQA/MOVDQU
867 // during the SSE domain pass.
868 let Predicates = [HasSSE1] in {
869 def : Pat<(alignedloadv4i32 addr:$src),
870 (MOVAPSrm addr:$src)>;
871 def : Pat<(loadv4i32 addr:$src),
872 (MOVUPSrm addr:$src)>;
873 def : Pat<(alignedloadv2i64 addr:$src),
874 (MOVAPSrm addr:$src)>;
875 def : Pat<(loadv2i64 addr:$src),
876 (MOVUPSrm addr:$src)>;
878 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
879 (MOVAPSmr addr:$dst, VR128:$src)>;
880 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
881 (MOVAPSmr addr:$dst, VR128:$src)>;
882 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
883 (MOVAPSmr addr:$dst, VR128:$src)>;
884 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
885 (MOVAPSmr addr:$dst, VR128:$src)>;
886 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
887 (MOVUPSmr addr:$dst, VR128:$src)>;
888 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
889 (MOVUPSmr addr:$dst, VR128:$src)>;
890 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
891 (MOVUPSmr addr:$dst, VR128:$src)>;
892 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
893 (MOVUPSmr addr:$dst, VR128:$src)>;
896 // Use vmovaps/vmovups for AVX integer load/store.
897 let Predicates = [HasAVX] in {
898 // 128-bit load/store
899 def : Pat<(alignedloadv4i32 addr:$src),
900 (VMOVAPSrm addr:$src)>;
901 def : Pat<(loadv4i32 addr:$src),
902 (VMOVUPSrm addr:$src)>;
903 def : Pat<(alignedloadv2i64 addr:$src),
904 (VMOVAPSrm addr:$src)>;
905 def : Pat<(loadv2i64 addr:$src),
906 (VMOVUPSrm addr:$src)>;
908 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
909 (VMOVAPSmr addr:$dst, VR128:$src)>;
910 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
911 (VMOVAPSmr addr:$dst, VR128:$src)>;
912 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
913 (VMOVAPSmr addr:$dst, VR128:$src)>;
914 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
915 (VMOVAPSmr addr:$dst, VR128:$src)>;
916 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
917 (VMOVUPSmr addr:$dst, VR128:$src)>;
918 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
919 (VMOVUPSmr addr:$dst, VR128:$src)>;
920 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
921 (VMOVUPSmr addr:$dst, VR128:$src)>;
922 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
923 (VMOVUPSmr addr:$dst, VR128:$src)>;
925 // 256-bit load/store
926 def : Pat<(alignedloadv4i64 addr:$src),
927 (VMOVAPSYrm addr:$src)>;
928 def : Pat<(loadv4i64 addr:$src),
929 (VMOVUPSYrm addr:$src)>;
930 def : Pat<(alignedloadv8i32 addr:$src),
931 (VMOVAPSYrm addr:$src)>;
932 def : Pat<(loadv8i32 addr:$src),
933 (VMOVUPSYrm addr:$src)>;
934 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
935 (VMOVAPSYmr addr:$dst, VR256:$src)>;
936 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
937 (VMOVAPSYmr addr:$dst, VR256:$src)>;
938 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
939 (VMOVAPSYmr addr:$dst, VR256:$src)>;
940 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
941 (VMOVAPSYmr addr:$dst, VR256:$src)>;
942 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
943 (VMOVUPSYmr addr:$dst, VR256:$src)>;
944 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
945 (VMOVUPSYmr addr:$dst, VR256:$src)>;
946 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
947 (VMOVUPSYmr addr:$dst, VR256:$src)>;
948 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
949 (VMOVUPSYmr addr:$dst, VR256:$src)>;
952 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
953 // bits are disregarded. FIXME: Set encoding to pseudo!
954 let neverHasSideEffects = 1 in {
955 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
956 "movaps\t{$src, $dst|$dst, $src}", []>;
957 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
958 "movapd\t{$src, $dst|$dst, $src}", []>;
959 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
960 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
961 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
962 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
965 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
966 // bits are disregarded. FIXME: Set encoding to pseudo!
967 let canFoldAsLoad = 1, isReMaterializable = 1 in {
968 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
969 "movaps\t{$src, $dst|$dst, $src}",
970 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
971 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
972 "movapd\t{$src, $dst|$dst, $src}",
973 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
974 let isCodeGenOnly = 1 in {
975 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
976 "movaps\t{$src, $dst|$dst, $src}",
977 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
978 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
979 "movapd\t{$src, $dst|$dst, $src}",
980 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
984 //===----------------------------------------------------------------------===//
985 // SSE 1 & 2 - Move Low packed FP Instructions
986 //===----------------------------------------------------------------------===//
988 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
989 PatFrag mov_frag, string base_opc,
991 def PSrm : PI<opc, MRMSrcMem,
992 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
993 !strconcat(base_opc, "s", asm_opr),
996 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
997 SSEPackedSingle>, TB;
999 def PDrm : PI<opc, MRMSrcMem,
1000 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1001 !strconcat(base_opc, "d", asm_opr),
1002 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1003 (scalar_to_vector (loadf64 addr:$src2)))))],
1004 SSEPackedDouble>, TB, OpSize;
1007 let AddedComplexity = 20 in {
1008 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1009 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1011 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1012 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1013 "\t{$src2, $dst|$dst, $src2}">;
1016 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1017 "movlps\t{$src, $dst|$dst, $src}",
1018 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1019 (iPTR 0))), addr:$dst)]>, VEX;
1020 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1021 "movlpd\t{$src, $dst|$dst, $src}",
1022 [(store (f64 (vector_extract (v2f64 VR128:$src),
1023 (iPTR 0))), addr:$dst)]>, VEX;
1024 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1025 "movlps\t{$src, $dst|$dst, $src}",
1026 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1027 (iPTR 0))), addr:$dst)]>;
1028 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1029 "movlpd\t{$src, $dst|$dst, $src}",
1030 [(store (f64 (vector_extract (v2f64 VR128:$src),
1031 (iPTR 0))), addr:$dst)]>;
1033 let Predicates = [HasAVX] in {
1034 let AddedComplexity = 20 in {
1035 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1036 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1037 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1038 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1039 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1040 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1041 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1042 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1043 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1044 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1047 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1048 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1049 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1050 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1051 VR128:$src2)), addr:$src1),
1052 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1054 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1055 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1056 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1057 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1058 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1060 // Shuffle with VMOVLPS
1061 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1062 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1063 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1064 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1065 def : Pat<(X86Movlps VR128:$src1,
1066 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1067 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1069 // Shuffle with VMOVLPD
1070 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1071 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1072 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1073 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1074 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1075 (scalar_to_vector (loadf64 addr:$src2)))),
1076 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1079 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1081 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1082 def : Pat<(store (v4i32 (X86Movlps
1083 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1084 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1085 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1087 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1088 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1090 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1093 let Predicates = [HasSSE1] in {
1094 let AddedComplexity = 20 in {
1095 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1096 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1097 (MOVLPSrm VR128:$src1, addr:$src2)>;
1098 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1099 (MOVLPSrm VR128:$src1, addr:$src2)>;
1102 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1103 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1104 (iPTR 0))), addr:$src1),
1105 (MOVLPSmr addr:$src1, VR128:$src2)>;
1106 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1107 (MOVLPSmr addr:$src1, VR128:$src2)>;
1108 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1109 VR128:$src2)), addr:$src1),
1110 (MOVLPSmr addr:$src1, VR128:$src2)>;
1112 // Shuffle with MOVLPS
1113 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1114 (MOVLPSrm VR128:$src1, addr:$src2)>;
1115 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1116 (MOVLPSrm VR128:$src1, addr:$src2)>;
1117 def : Pat<(X86Movlps VR128:$src1,
1118 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1119 (MOVLPSrm VR128:$src1, addr:$src2)>;
1120 def : Pat<(X86Movlps VR128:$src1,
1121 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1122 (MOVLPSrm VR128:$src1, addr:$src2)>;
1125 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1127 (MOVLPSmr addr:$src1, VR128:$src2)>;
1128 def : Pat<(store (v4i32 (X86Movlps
1129 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1131 (MOVLPSmr addr:$src1, VR128:$src2)>;
1134 let Predicates = [HasSSE2] in {
1135 let AddedComplexity = 20 in {
1136 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1137 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1138 (MOVLPDrm VR128:$src1, addr:$src2)>;
1139 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1140 (MOVLPDrm VR128:$src1, addr:$src2)>;
1143 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1144 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1145 (MOVLPDmr addr:$src1, VR128:$src2)>;
1146 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1147 (MOVLPDmr addr:$src1, VR128:$src2)>;
1149 // Shuffle with MOVLPD
1150 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1151 (MOVLPDrm VR128:$src1, addr:$src2)>;
1152 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1153 (MOVLPDrm VR128:$src1, addr:$src2)>;
1154 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1155 (scalar_to_vector (loadf64 addr:$src2)))),
1156 (MOVLPDrm VR128:$src1, addr:$src2)>;
1159 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1161 (MOVLPDmr addr:$src1, VR128:$src2)>;
1162 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1164 (MOVLPDmr addr:$src1, VR128:$src2)>;
1167 //===----------------------------------------------------------------------===//
1168 // SSE 1 & 2 - Move Hi packed FP Instructions
1169 //===----------------------------------------------------------------------===//
1171 let AddedComplexity = 20 in {
1172 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1173 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1175 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1176 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1177 "\t{$src2, $dst|$dst, $src2}">;
1180 // v2f64 extract element 1 is always custom lowered to unpack high to low
1181 // and extract element 0 so the non-store version isn't too horrible.
1182 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1183 "movhps\t{$src, $dst|$dst, $src}",
1184 [(store (f64 (vector_extract
1185 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1186 (undef)), (iPTR 0))), addr:$dst)]>,
1188 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1189 "movhpd\t{$src, $dst|$dst, $src}",
1190 [(store (f64 (vector_extract
1191 (v2f64 (unpckh VR128:$src, (undef))),
1192 (iPTR 0))), addr:$dst)]>,
1194 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1195 "movhps\t{$src, $dst|$dst, $src}",
1196 [(store (f64 (vector_extract
1197 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1198 (undef)), (iPTR 0))), addr:$dst)]>;
1199 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1200 "movhpd\t{$src, $dst|$dst, $src}",
1201 [(store (f64 (vector_extract
1202 (v2f64 (unpckh VR128:$src, (undef))),
1203 (iPTR 0))), addr:$dst)]>;
1205 let Predicates = [HasAVX] in {
1207 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1208 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1209 def : Pat<(X86Movlhps VR128:$src1,
1210 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1211 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1212 def : Pat<(X86Movlhps VR128:$src1,
1213 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1214 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1215 def : Pat<(X86Movlhps VR128:$src1,
1216 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1217 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1219 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1220 // is during lowering, where it's not possible to recognize the load fold cause
1221 // it has two uses through a bitcast. One use disappears at isel time and the
1222 // fold opportunity reappears.
1223 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1224 (scalar_to_vector (loadf64 addr:$src2)))),
1225 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1227 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1228 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1229 (scalar_to_vector (loadf64 addr:$src2)))),
1230 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1233 def : Pat<(store (f64 (vector_extract
1234 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1235 (VMOVHPSmr addr:$dst, VR128:$src)>;
1236 def : Pat<(store (f64 (vector_extract
1237 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1238 (VMOVHPDmr addr:$dst, VR128:$src)>;
1241 let Predicates = [HasSSE1] in {
1243 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1244 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1245 def : Pat<(X86Movlhps VR128:$src1,
1246 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1247 (MOVHPSrm VR128:$src1, addr:$src2)>;
1248 def : Pat<(X86Movlhps VR128:$src1,
1249 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1250 (MOVHPSrm VR128:$src1, addr:$src2)>;
1251 def : Pat<(X86Movlhps VR128:$src1,
1252 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1253 (MOVHPSrm VR128:$src1, addr:$src2)>;
1256 def : Pat<(store (f64 (vector_extract
1257 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1258 (MOVHPSmr addr:$dst, VR128:$src)>;
1261 let Predicates = [HasSSE2] in {
1262 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1263 // is during lowering, where it's not possible to recognize the load fold cause
1264 // it has two uses through a bitcast. One use disappears at isel time and the
1265 // fold opportunity reappears.
1266 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1267 (scalar_to_vector (loadf64 addr:$src2)))),
1268 (MOVHPDrm VR128:$src1, addr:$src2)>;
1270 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1271 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1272 (scalar_to_vector (loadf64 addr:$src2)))),
1273 (MOVHPDrm VR128:$src1, addr:$src2)>;
1276 def : Pat<(store (f64 (vector_extract
1277 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1278 (MOVHPDmr addr:$dst, VR128:$src)>;
1281 //===----------------------------------------------------------------------===//
1282 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1283 //===----------------------------------------------------------------------===//
1285 let AddedComplexity = 20 in {
1286 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1287 (ins VR128:$src1, VR128:$src2),
1288 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1290 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1292 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1293 (ins VR128:$src1, VR128:$src2),
1294 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1296 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1299 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1300 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1301 (ins VR128:$src1, VR128:$src2),
1302 "movlhps\t{$src2, $dst|$dst, $src2}",
1304 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1305 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1306 (ins VR128:$src1, VR128:$src2),
1307 "movhlps\t{$src2, $dst|$dst, $src2}",
1309 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1312 let Predicates = [HasAVX] in {
1314 let AddedComplexity = 20 in {
1315 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1316 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1317 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1318 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1320 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1321 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1322 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1324 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1325 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1326 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1327 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1328 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1329 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1332 let AddedComplexity = 20 in {
1333 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1334 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1335 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1337 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1338 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1339 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1340 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1341 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1344 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1345 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1346 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1347 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1350 let Predicates = [HasSSE1] in {
1352 let AddedComplexity = 20 in {
1353 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1354 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1355 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1356 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1358 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1359 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1360 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1362 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1363 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1364 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1365 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1366 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1367 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1370 let AddedComplexity = 20 in {
1371 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1372 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1373 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1375 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1376 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1377 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1378 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1379 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1382 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1383 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1384 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1385 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1388 //===----------------------------------------------------------------------===//
1389 // SSE 1 & 2 - Conversion Instructions
1390 //===----------------------------------------------------------------------===//
1392 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1393 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1395 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1396 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1397 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1398 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1401 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1402 X86MemOperand x86memop, string asm> {
1403 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1405 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1408 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1409 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1410 string asm, Domain d> {
1411 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1412 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1413 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1414 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1417 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1418 X86MemOperand x86memop, string asm> {
1419 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1420 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1422 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1423 (ins DstRC:$src1, x86memop:$src),
1424 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1427 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1428 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1430 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1431 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1433 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1434 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1436 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1437 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1438 VEX, VEX_W, VEX_LIG;
1440 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1441 // register, but the same isn't true when only using memory operands,
1442 // provide other assembly "l" and "q" forms to address this explicitly
1443 // where appropriate to do so.
1444 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1446 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1447 VEX_4V, VEX_W, VEX_LIG;
1448 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1450 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1452 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1453 VEX_4V, VEX_W, VEX_LIG;
1455 let Predicates = [HasAVX] in {
1456 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1457 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1458 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1459 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1460 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1461 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1462 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1463 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f32 (sint_to_fp GR32:$src)),
1466 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1467 def : Pat<(f32 (sint_to_fp GR64:$src)),
1468 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1469 def : Pat<(f64 (sint_to_fp GR32:$src)),
1470 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1471 def : Pat<(f64 (sint_to_fp GR64:$src)),
1472 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1475 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1476 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1477 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1478 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1479 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1480 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1481 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1482 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1483 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1484 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1485 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1486 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1487 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1488 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1489 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1490 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1492 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1493 // and/or XMM operand(s).
1495 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1496 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1498 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1499 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1500 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1501 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1502 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1503 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1506 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1507 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1508 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1509 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1511 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1512 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1513 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1514 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1515 (ins DstRC:$src1, x86memop:$src2),
1517 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1518 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1519 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1522 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1523 f128mem, load, "cvtsd2si">, XD, VEX;
1524 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1525 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1528 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1529 // Get rid of this hack or rename the intrinsics, there are several
1530 // intructions that only match with the intrinsic form, why create duplicates
1531 // to let them be recognized by the assembler?
1532 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1533 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1534 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1535 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1538 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1539 f128mem, load, "cvtsd2si{l}">, XD;
1540 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1541 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1544 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1545 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1546 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1547 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1549 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1550 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1551 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1552 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1555 let Constraints = "$src1 = $dst" in {
1556 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1557 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1559 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1560 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1561 "cvtsi2ss{q}">, XS, REX_W;
1562 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1563 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1565 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1566 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1567 "cvtsi2sd">, XD, REX_W;
1572 // Aliases for intrinsics
1573 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1574 f32mem, load, "cvttss2si">, XS, VEX;
1575 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1576 int_x86_sse_cvttss2si64, f32mem, load,
1577 "cvttss2si">, XS, VEX, VEX_W;
1578 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1579 f128mem, load, "cvttsd2si">, XD, VEX;
1580 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1581 int_x86_sse2_cvttsd2si64, f128mem, load,
1582 "cvttsd2si">, XD, VEX, VEX_W;
1583 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1584 f32mem, load, "cvttss2si">, XS;
1585 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1586 int_x86_sse_cvttss2si64, f32mem, load,
1587 "cvttss2si{q}">, XS, REX_W;
1588 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1589 f128mem, load, "cvttsd2si">, XD;
1590 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1591 int_x86_sse2_cvttsd2si64, f128mem, load,
1592 "cvttsd2si{q}">, XD, REX_W;
1594 let Pattern = []<dag> in {
1595 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1596 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1598 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1599 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1601 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1602 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1603 SSEPackedSingle>, TB, VEX;
1604 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1605 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1606 SSEPackedSingle>, TB, VEX;
1609 let Pattern = []<dag> in {
1610 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1611 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1612 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1613 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1614 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1615 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1616 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1619 let Predicates = [HasSSE1] in {
1620 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1621 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1622 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1623 (CVTSS2SIrm addr:$src)>;
1624 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1625 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1626 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1627 (CVTSS2SI64rm addr:$src)>;
1630 let Predicates = [HasAVX] in {
1631 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1632 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1633 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1634 (VCVTSS2SIrm addr:$src)>;
1635 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1636 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1637 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1638 (VCVTSS2SI64rm addr:$src)>;
1643 // Convert scalar double to scalar single
1644 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1645 (ins FR64:$src1, FR64:$src2),
1646 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1649 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1650 (ins FR64:$src1, f64mem:$src2),
1651 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1652 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1654 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1657 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1658 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1659 [(set FR32:$dst, (fround FR64:$src))]>;
1660 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1661 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1662 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1663 Requires<[HasSSE2, OptForSize]>;
1665 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1666 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1668 let Constraints = "$src1 = $dst" in
1669 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1670 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1672 // Convert scalar single to scalar double
1673 // SSE2 instructions with XS prefix
1674 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1675 (ins FR32:$src1, FR32:$src2),
1676 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1677 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1679 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1680 (ins FR32:$src1, f32mem:$src2),
1681 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1682 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1684 let Predicates = [HasAVX] in {
1685 def : Pat<(f64 (fextend FR32:$src)),
1686 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1687 def : Pat<(fextend (loadf32 addr:$src)),
1688 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1689 def : Pat<(extloadf32 addr:$src),
1690 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1693 def : Pat<(extloadf32 addr:$src),
1694 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1695 Requires<[HasAVX, OptForSpeed]>;
1697 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1698 "cvtss2sd\t{$src, $dst|$dst, $src}",
1699 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1700 Requires<[HasSSE2]>;
1701 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1702 "cvtss2sd\t{$src, $dst|$dst, $src}",
1703 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1704 Requires<[HasSSE2, OptForSize]>;
1706 // extload f32 -> f64. This matches load+fextend because we have a hack in
1707 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1709 // Since these loads aren't folded into the fextend, we have to match it
1711 def : Pat<(fextend (loadf32 addr:$src)),
1712 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1713 def : Pat<(extloadf32 addr:$src),
1714 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1716 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1717 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1718 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1719 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1720 VR128:$src2))]>, XS, VEX_4V,
1722 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1723 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1724 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1725 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1726 (load addr:$src2)))]>, XS, VEX_4V,
1728 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1729 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1730 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1731 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1732 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1733 VR128:$src2))]>, XS,
1734 Requires<[HasSSE2]>;
1735 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1736 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1737 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1738 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1739 (load addr:$src2)))]>, XS,
1740 Requires<[HasSSE2]>;
1743 // Convert doubleword to packed single/double fp
1744 // SSE2 instructions without OpSize prefix
1745 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1746 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1747 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1748 TB, VEX, Requires<[HasAVX]>;
1749 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1750 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1751 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1752 (bitconvert (memopv2i64 addr:$src))))]>,
1753 TB, VEX, Requires<[HasAVX]>;
1754 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1755 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1756 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1757 TB, Requires<[HasSSE2]>;
1758 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1759 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1760 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1761 (bitconvert (memopv2i64 addr:$src))))]>,
1762 TB, Requires<[HasSSE2]>;
1764 // FIXME: why the non-intrinsic version is described as SSE3?
1765 // SSE2 instructions with XS prefix
1766 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1767 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1768 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1769 XS, VEX, Requires<[HasAVX]>;
1770 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1771 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1772 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1773 (bitconvert (memopv2i64 addr:$src))))]>,
1774 XS, VEX, Requires<[HasAVX]>;
1775 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1776 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1777 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1778 XS, Requires<[HasSSE2]>;
1779 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1780 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1781 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1782 (bitconvert (memopv2i64 addr:$src))))]>,
1783 XS, Requires<[HasSSE2]>;
1786 // Convert packed single/double fp to doubleword
1787 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1788 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1789 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1790 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1791 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1792 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1793 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1794 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1795 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1796 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1797 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1798 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1800 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1801 "cvtps2dq\t{$src, $dst|$dst, $src}",
1802 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1804 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1806 "cvtps2dq\t{$src, $dst|$dst, $src}",
1807 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1808 (memop addr:$src)))]>, VEX;
1809 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1810 "cvtps2dq\t{$src, $dst|$dst, $src}",
1811 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1812 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1813 "cvtps2dq\t{$src, $dst|$dst, $src}",
1814 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1815 (memop addr:$src)))]>;
1817 // SSE2 packed instructions with XD prefix
1818 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1819 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1820 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1821 XD, VEX, Requires<[HasAVX]>;
1822 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1823 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1824 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1825 (memop addr:$src)))]>,
1826 XD, VEX, Requires<[HasAVX]>;
1827 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1828 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1829 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1830 XD, Requires<[HasSSE2]>;
1831 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1832 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1833 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1834 (memop addr:$src)))]>,
1835 XD, Requires<[HasSSE2]>;
1838 // Convert with truncation packed single/double fp to doubleword
1839 // SSE2 packed instructions with XS prefix
1840 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1841 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1843 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1844 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1845 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1848 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1849 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1850 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1851 "cvttps2dq\t{$src, $dst|$dst, $src}",
1853 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1854 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1855 "cvttps2dq\t{$src, $dst|$dst, $src}",
1857 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1859 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1860 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1862 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1863 XS, VEX, Requires<[HasAVX]>;
1864 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1865 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1866 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1867 (memop addr:$src)))]>,
1868 XS, VEX, Requires<[HasAVX]>;
1870 let Predicates = [HasSSE2] in {
1871 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1872 (Int_CVTDQ2PSrr VR128:$src)>;
1873 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1874 (CVTTPS2DQrr VR128:$src)>;
1877 let Predicates = [HasAVX] in {
1878 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1879 (Int_VCVTDQ2PSrr VR128:$src)>;
1880 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1881 (VCVTTPS2DQrr VR128:$src)>;
1882 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1883 (VCVTDQ2PSYrr VR256:$src)>;
1884 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1885 (VCVTTPS2DQYrr VR256:$src)>;
1888 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1889 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1891 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1892 let isCodeGenOnly = 1 in
1893 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1894 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1895 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1896 (memop addr:$src)))]>, VEX;
1897 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1898 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1899 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1900 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1901 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1902 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1903 (memop addr:$src)))]>;
1905 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1906 // register, but the same isn't true when using memory operands instead.
1907 // Provide other assembly rr and rm forms to address this explicitly.
1908 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1909 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1912 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1913 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1914 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1915 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1918 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1919 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1920 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1921 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1923 // Convert packed single to packed double
1924 let Predicates = [HasAVX] in {
1925 // SSE2 instructions without OpSize prefix
1926 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1927 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1928 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1929 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1930 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1931 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1932 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1933 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1935 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1936 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1937 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1938 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1940 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1941 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1942 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1943 TB, VEX, Requires<[HasAVX]>;
1944 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1945 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1946 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1947 (load addr:$src)))]>,
1948 TB, VEX, Requires<[HasAVX]>;
1949 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1950 "cvtps2pd\t{$src, $dst|$dst, $src}",
1951 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1952 TB, Requires<[HasSSE2]>;
1953 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1954 "cvtps2pd\t{$src, $dst|$dst, $src}",
1955 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1956 (load addr:$src)))]>,
1957 TB, Requires<[HasSSE2]>;
1959 // Convert packed double to packed single
1960 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1961 // register, but the same isn't true when using memory operands instead.
1962 // Provide other assembly rr and rm forms to address this explicitly.
1963 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1964 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1965 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1966 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1969 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1970 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1971 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1972 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1975 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1976 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1977 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1978 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1979 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1980 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1981 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1982 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1985 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1986 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1987 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1988 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1990 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1991 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1992 (memop addr:$src)))]>;
1993 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1994 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1995 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1996 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1997 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1998 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1999 (memop addr:$src)))]>;
2001 // AVX 256-bit register conversion intrinsics
2002 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2003 // whenever possible to avoid declaring two versions of each one.
2004 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2005 (VCVTDQ2PSYrr VR256:$src)>;
2006 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2007 (VCVTDQ2PSYrm addr:$src)>;
2009 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2010 (VCVTPD2PSYrr VR256:$src)>;
2011 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2012 (VCVTPD2PSYrm addr:$src)>;
2014 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2015 (VCVTPS2DQYrr VR256:$src)>;
2016 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2017 (VCVTPS2DQYrm addr:$src)>;
2019 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2020 (VCVTPS2PDYrr VR128:$src)>;
2021 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2022 (VCVTPS2PDYrm addr:$src)>;
2024 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2025 (VCVTTPD2DQYrr VR256:$src)>;
2026 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2027 (VCVTTPD2DQYrm addr:$src)>;
2029 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2030 (VCVTTPS2DQYrr VR256:$src)>;
2031 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2032 (VCVTTPS2DQYrm addr:$src)>;
2034 // Match fround and fextend for 128/256-bit conversions
2035 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2036 (VCVTPD2PSYrr VR256:$src)>;
2037 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2038 (VCVTPD2PSYrm addr:$src)>;
2040 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2041 (VCVTPS2PDYrr VR128:$src)>;
2042 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2043 (VCVTPS2PDYrm addr:$src)>;
2045 //===----------------------------------------------------------------------===//
2046 // SSE 1 & 2 - Compare Instructions
2047 //===----------------------------------------------------------------------===//
2049 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2050 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2051 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2052 string asm, string asm_alt> {
2053 def rr : SIi8<0xC2, MRMSrcReg,
2054 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2055 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2056 def rm : SIi8<0xC2, MRMSrcMem,
2057 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2058 [(set RC:$dst, (OpNode (VT RC:$src1),
2059 (ld_frag addr:$src2), imm:$cc))]>;
2061 // Accept explicit immediate argument form instead of comparison code.
2062 let neverHasSideEffects = 1 in {
2063 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2064 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2066 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2067 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2071 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2072 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2073 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2074 XS, VEX_4V, VEX_LIG;
2075 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2076 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2077 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2078 XD, VEX_4V, VEX_LIG;
2080 let Constraints = "$src1 = $dst" in {
2081 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2082 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2083 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2085 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2086 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2087 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2091 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2092 Intrinsic Int, string asm> {
2093 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2094 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2095 [(set VR128:$dst, (Int VR128:$src1,
2096 VR128:$src, imm:$cc))]>;
2097 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2098 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2099 [(set VR128:$dst, (Int VR128:$src1,
2100 (load addr:$src), imm:$cc))]>;
2103 // Aliases to match intrinsics which expect XMM operand(s).
2104 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2105 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2107 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2108 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2110 let Constraints = "$src1 = $dst" in {
2111 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2112 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2113 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2114 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2118 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2119 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2120 ValueType vt, X86MemOperand x86memop,
2121 PatFrag ld_frag, string OpcodeStr, Domain d> {
2122 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2123 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2124 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2125 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2126 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2127 [(set EFLAGS, (OpNode (vt RC:$src1),
2128 (ld_frag addr:$src2)))], d>;
2131 let Defs = [EFLAGS] in {
2132 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2133 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2134 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2135 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2137 let Pattern = []<dag> in {
2138 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2139 "comiss", SSEPackedSingle>, TB, VEX,
2141 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2142 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2146 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2147 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2148 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2149 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2151 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2152 load, "comiss", SSEPackedSingle>, TB, VEX;
2153 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2154 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2155 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2156 "ucomiss", SSEPackedSingle>, TB;
2157 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2158 "ucomisd", SSEPackedDouble>, TB, OpSize;
2160 let Pattern = []<dag> in {
2161 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2162 "comiss", SSEPackedSingle>, TB;
2163 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2164 "comisd", SSEPackedDouble>, TB, OpSize;
2167 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2168 load, "ucomiss", SSEPackedSingle>, TB;
2169 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2170 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2172 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2173 "comiss", SSEPackedSingle>, TB;
2174 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2175 "comisd", SSEPackedDouble>, TB, OpSize;
2176 } // Defs = [EFLAGS]
2178 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2179 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2180 Intrinsic Int, string asm, string asm_alt,
2182 let isAsmParserOnly = 1 in {
2183 def rri : PIi8<0xC2, MRMSrcReg,
2184 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2185 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2186 def rmi : PIi8<0xC2, MRMSrcMem,
2187 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2188 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2191 // Accept explicit immediate argument form instead of comparison code.
2192 def rri_alt : PIi8<0xC2, MRMSrcReg,
2193 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2195 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2196 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2200 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2201 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2202 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2203 SSEPackedSingle>, TB, VEX_4V;
2204 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2205 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2206 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2207 SSEPackedDouble>, TB, OpSize, VEX_4V;
2208 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2209 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2210 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2211 SSEPackedSingle>, TB, VEX_4V;
2212 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2213 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2214 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2215 SSEPackedDouble>, TB, OpSize, VEX_4V;
2216 let Constraints = "$src1 = $dst" in {
2217 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2218 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2219 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2220 SSEPackedSingle>, TB;
2221 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2222 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2223 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2224 SSEPackedDouble>, TB, OpSize;
2227 let Predicates = [HasSSE1] in {
2228 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2229 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2230 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2231 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2234 let Predicates = [HasSSE2] in {
2235 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2236 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2237 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2238 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2241 let Predicates = [HasAVX] in {
2242 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2243 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2244 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2245 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2246 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2247 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2248 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2249 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2251 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2252 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2253 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2254 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2255 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2256 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2257 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2258 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2261 //===----------------------------------------------------------------------===//
2262 // SSE 1 & 2 - Shuffle Instructions
2263 //===----------------------------------------------------------------------===//
2265 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2266 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2267 ValueType vt, string asm, PatFrag mem_frag,
2268 Domain d, bit IsConvertibleToThreeAddress = 0> {
2269 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2270 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2271 [(set RC:$dst, (vt (shufp:$src3
2272 RC:$src1, (mem_frag addr:$src2))))], d>;
2273 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2274 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2275 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2277 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2280 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2281 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2282 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2283 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2284 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2285 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2286 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2287 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2288 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2289 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2290 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2291 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2293 let Constraints = "$src1 = $dst" in {
2294 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2295 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2296 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2298 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2299 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2300 memopv2f64, SSEPackedDouble>, TB, OpSize;
2303 let Predicates = [HasSSE1] in {
2304 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2305 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2306 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2307 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2308 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2309 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2310 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2311 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2312 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2313 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2314 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2315 // fall back to this for SSE1)
2316 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2317 (SHUFPSrri VR128:$src2, VR128:$src1,
2318 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2319 // Special unary SHUFPSrri case.
2320 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2321 (SHUFPSrri VR128:$src1, VR128:$src1,
2322 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2325 let Predicates = [HasSSE2] in {
2326 // Special binary v4i32 shuffle cases with SHUFPS.
2327 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2328 (SHUFPSrri VR128:$src1, VR128:$src2,
2329 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2331 (bc_v4i32 (memopv2i64 addr:$src2)))),
2332 (SHUFPSrmi VR128:$src1, addr:$src2,
2333 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2334 // Special unary SHUFPDrri cases.
2335 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2336 (SHUFPDrri VR128:$src1, VR128:$src1,
2337 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2338 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2339 (SHUFPDrri VR128:$src1, VR128:$src1,
2340 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2341 // Special binary v2i64 shuffle cases using SHUFPDrri.
2342 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2343 (SHUFPDrri VR128:$src1, VR128:$src2,
2344 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2345 // Generic SHUFPD patterns
2346 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2347 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2348 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2349 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2350 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2351 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2352 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2355 let Predicates = [HasAVX] in {
2356 def : Pat<(v4f32 (X86Shufps VR128:$src1,
2357 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2358 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2359 def : Pat<(v4f32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2360 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2361 def : Pat<(v4i32 (X86Shufps VR128:$src1,
2362 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2363 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2364 def : Pat<(v4i32 (X86Shufps VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2365 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2366 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2367 // fall back to this for SSE1)
2368 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2369 (VSHUFPSrri VR128:$src2, VR128:$src1,
2370 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2371 // Special unary SHUFPSrri case.
2372 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2373 (VSHUFPSrri VR128:$src1, VR128:$src1,
2374 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2375 // Special binary v4i32 shuffle cases with SHUFPS.
2376 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2377 (VSHUFPSrri VR128:$src1, VR128:$src2,
2378 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2379 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2380 (bc_v4i32 (memopv2i64 addr:$src2)))),
2381 (VSHUFPSrmi VR128:$src1, addr:$src2,
2382 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2383 // Special unary SHUFPDrri cases.
2384 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2385 (VSHUFPDrri VR128:$src1, VR128:$src1,
2386 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2387 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2388 (VSHUFPDrri VR128:$src1, VR128:$src1,
2389 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2390 // Special binary v2i64 shuffle cases using SHUFPDrri.
2391 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2392 (VSHUFPDrri VR128:$src1, VR128:$src2,
2393 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2395 def : Pat<(v2f64 (X86Shufps VR128:$src1,
2396 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2397 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2398 def : Pat<(v2i64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2399 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2400 def : Pat<(v2f64 (X86Shufpd VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2401 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2404 def : Pat<(v8i32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2405 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2406 def : Pat<(v8i32 (X86Shufps VR256:$src1,
2407 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2408 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2410 def : Pat<(v8f32 (X86Shufps VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2411 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2412 def : Pat<(v8f32 (X86Shufps VR256:$src1,
2413 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2414 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2416 def : Pat<(v4i64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2417 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2418 def : Pat<(v4i64 (X86Shufpd VR256:$src1,
2419 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2420 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2422 def : Pat<(v4f64 (X86Shufpd VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2423 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2424 def : Pat<(v4f64 (X86Shufpd VR256:$src1,
2425 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2426 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2429 //===----------------------------------------------------------------------===//
2430 // SSE 1 & 2 - Unpack Instructions
2431 //===----------------------------------------------------------------------===//
2433 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2434 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2435 PatFrag mem_frag, RegisterClass RC,
2436 X86MemOperand x86memop, string asm,
2438 def rr : PI<opc, MRMSrcReg,
2439 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2441 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2442 def rm : PI<opc, MRMSrcMem,
2443 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2445 (vt (OpNode RC:$src1,
2446 (mem_frag addr:$src2))))], d>;
2449 let AddedComplexity = 10 in {
2450 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2451 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2452 SSEPackedSingle>, TB, VEX_4V;
2453 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2454 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2455 SSEPackedDouble>, TB, OpSize, VEX_4V;
2456 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2457 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2458 SSEPackedSingle>, TB, VEX_4V;
2459 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2460 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2461 SSEPackedDouble>, TB, OpSize, VEX_4V;
2463 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2464 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2465 SSEPackedSingle>, TB, VEX_4V;
2466 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2467 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2468 SSEPackedDouble>, TB, OpSize, VEX_4V;
2469 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2470 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2471 SSEPackedSingle>, TB, VEX_4V;
2472 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2473 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2474 SSEPackedDouble>, TB, OpSize, VEX_4V;
2476 let Constraints = "$src1 = $dst" in {
2477 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2478 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2479 SSEPackedSingle>, TB;
2480 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2481 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2482 SSEPackedDouble>, TB, OpSize;
2483 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2484 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2485 SSEPackedSingle>, TB;
2486 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2487 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2488 SSEPackedDouble>, TB, OpSize;
2489 } // Constraints = "$src1 = $dst"
2490 } // AddedComplexity
2492 let Predicates = [HasSSE1] in {
2493 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2494 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2495 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2496 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2497 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2498 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2499 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2500 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2503 let Predicates = [HasSSE2] in {
2504 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2505 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2506 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2507 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2508 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2509 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2510 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2511 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2513 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2514 // problem is during lowering, where it's not possible to recognize the load
2515 // fold cause it has two uses through a bitcast. One use disappears at isel
2516 // time and the fold opportunity reappears.
2517 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2518 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2520 let AddedComplexity = 10 in
2521 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2522 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2525 let Predicates = [HasAVX] in {
2526 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2527 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2528 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2529 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2530 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2531 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2532 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2533 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2535 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2536 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2537 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2538 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2539 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2540 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2541 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2542 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2544 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2545 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2546 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2547 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2548 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2549 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2550 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2551 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2553 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2554 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2555 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2556 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2557 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2558 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2559 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2560 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2562 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2563 // problem is during lowering, where it's not possible to recognize the load
2564 // fold cause it has two uses through a bitcast. One use disappears at isel
2565 // time and the fold opportunity reappears.
2566 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2567 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2568 let AddedComplexity = 10 in
2569 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2570 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2573 //===----------------------------------------------------------------------===//
2574 // SSE 1 & 2 - Extract Floating-Point Sign mask
2575 //===----------------------------------------------------------------------===//
2577 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2578 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2580 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2581 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2582 [(set GR32:$dst, (Int RC:$src))], d>;
2583 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2584 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2587 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2588 SSEPackedSingle>, TB;
2589 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2590 SSEPackedDouble>, TB, OpSize;
2592 def : Pat<(i32 (X86fgetsign FR32:$src)),
2593 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2594 sub_ss))>, Requires<[HasSSE1]>;
2595 def : Pat<(i64 (X86fgetsign FR32:$src)),
2596 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2597 sub_ss))>, Requires<[HasSSE1]>;
2598 def : Pat<(i32 (X86fgetsign FR64:$src)),
2599 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2600 sub_sd))>, Requires<[HasSSE2]>;
2601 def : Pat<(i64 (X86fgetsign FR64:$src)),
2602 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2603 sub_sd))>, Requires<[HasSSE2]>;
2605 let Predicates = [HasAVX] in {
2606 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2607 "movmskps", SSEPackedSingle>, TB, VEX;
2608 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2609 "movmskpd", SSEPackedDouble>, TB,
2611 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2612 "movmskps", SSEPackedSingle>, TB, VEX;
2613 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2614 "movmskpd", SSEPackedDouble>, TB,
2617 def : Pat<(i32 (X86fgetsign FR32:$src)),
2618 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2620 def : Pat<(i64 (X86fgetsign FR32:$src)),
2621 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2623 def : Pat<(i32 (X86fgetsign FR64:$src)),
2624 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2626 def : Pat<(i64 (X86fgetsign FR64:$src)),
2627 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2631 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2632 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2633 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2634 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2636 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2637 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2638 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2639 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2643 //===----------------------------------------------------------------------===//
2644 // SSE 1 & 2 - Logical Instructions
2645 //===----------------------------------------------------------------------===//
2647 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2649 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2651 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2652 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2654 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2655 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2657 let Constraints = "$src1 = $dst" in {
2658 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2659 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2661 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2662 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2666 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2667 let mayLoad = 0 in {
2668 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2669 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2670 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2673 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2674 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2676 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2678 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2680 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2681 // are all promoted to v2i64, and the patterns are covered by the int
2682 // version. This is needed in SSE only, because v2i64 isn't supported on
2683 // SSE1, but only on SSE2.
2684 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2685 !strconcat(OpcodeStr, "ps"), f128mem, [],
2686 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2687 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2689 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2690 !strconcat(OpcodeStr, "pd"), f128mem,
2691 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2692 (bc_v2i64 (v2f64 VR128:$src2))))],
2693 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2694 (memopv2i64 addr:$src2)))], 0>,
2696 let Constraints = "$src1 = $dst" in {
2697 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2698 !strconcat(OpcodeStr, "ps"), f128mem,
2699 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2700 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2701 (memopv2i64 addr:$src2)))]>, TB;
2703 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2704 !strconcat(OpcodeStr, "pd"), f128mem,
2705 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2706 (bc_v2i64 (v2f64 VR128:$src2))))],
2707 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2708 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2712 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2714 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2716 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2717 !strconcat(OpcodeStr, "ps"), f256mem,
2718 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2719 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2720 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2722 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2723 !strconcat(OpcodeStr, "pd"), f256mem,
2724 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2725 (bc_v4i64 (v4f64 VR256:$src2))))],
2726 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2727 (memopv4i64 addr:$src2)))], 0>,
2731 // AVX 256-bit packed logical ops forms
2732 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2733 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2734 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2735 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2737 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2738 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2739 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2740 let isCommutable = 0 in
2741 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2743 //===----------------------------------------------------------------------===//
2744 // SSE 1 & 2 - Arithmetic Instructions
2745 //===----------------------------------------------------------------------===//
2747 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2750 /// In addition, we also have a special variant of the scalar form here to
2751 /// represent the associated intrinsic operation. This form is unlike the
2752 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2753 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2755 /// These three forms can each be reg+reg or reg+mem.
2758 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2760 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2762 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2763 OpNode, FR32, f32mem, Is2Addr>, XS;
2764 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2765 OpNode, FR64, f64mem, Is2Addr>, XD;
2768 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2770 let mayLoad = 0 in {
2771 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2772 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2773 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2774 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2778 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2780 let mayLoad = 0 in {
2781 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2782 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2783 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2784 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2788 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2790 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2791 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2792 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2793 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2796 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2798 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2799 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2800 SSEPackedSingle, Is2Addr>, TB;
2802 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2803 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2804 SSEPackedDouble, Is2Addr>, TB, OpSize;
2807 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2808 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2809 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2810 SSEPackedSingle, 0>, TB;
2812 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2813 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2814 SSEPackedDouble, 0>, TB, OpSize;
2817 // Binary Arithmetic instructions
2818 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2819 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2820 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2821 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2822 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2823 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2824 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2825 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2827 let isCommutable = 0 in {
2828 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2829 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2830 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2831 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2832 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2833 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2834 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2835 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2836 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2837 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2838 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2839 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2840 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2841 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2842 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2843 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2844 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2845 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2846 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2847 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2850 let Constraints = "$src1 = $dst" in {
2851 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2852 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2853 basic_sse12_fp_binop_s_int<0x58, "add">;
2854 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2855 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2856 basic_sse12_fp_binop_s_int<0x59, "mul">;
2858 let isCommutable = 0 in {
2859 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2860 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2861 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2862 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2863 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2864 basic_sse12_fp_binop_s_int<0x5E, "div">;
2865 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2866 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2867 basic_sse12_fp_binop_s_int<0x5F, "max">,
2868 basic_sse12_fp_binop_p_int<0x5F, "max">;
2869 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2870 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2871 basic_sse12_fp_binop_s_int<0x5D, "min">,
2872 basic_sse12_fp_binop_p_int<0x5D, "min">;
2877 /// In addition, we also have a special variant of the scalar form here to
2878 /// represent the associated intrinsic operation. This form is unlike the
2879 /// plain scalar form, in that it takes an entire vector (instead of a
2880 /// scalar) and leaves the top elements undefined.
2882 /// And, we have a special variant form for a full-vector intrinsic form.
2884 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2885 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2886 SDNode OpNode, Intrinsic F32Int> {
2887 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2888 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2889 [(set FR32:$dst, (OpNode FR32:$src))]>;
2890 // For scalar unary operations, fold a load into the operation
2891 // only in OptForSize mode. It eliminates an instruction, but it also
2892 // eliminates a whole-register clobber (the load), so it introduces a
2893 // partial register update condition.
2894 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2895 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2896 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2897 Requires<[HasSSE1, OptForSize]>;
2898 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2899 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2900 [(set VR128:$dst, (F32Int VR128:$src))]>;
2901 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2902 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2903 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2906 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2907 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2908 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2909 !strconcat(OpcodeStr,
2910 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2912 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2913 !strconcat(OpcodeStr,
2914 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2915 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2916 (ins VR128:$src1, ssmem:$src2),
2917 !strconcat(OpcodeStr,
2918 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2921 /// sse1_fp_unop_p - SSE1 unops in packed form.
2922 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2923 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2924 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2925 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2926 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2927 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2928 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2931 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2932 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2933 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2934 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2935 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2936 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2937 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2938 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2941 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2942 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2943 Intrinsic V4F32Int> {
2944 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2945 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2946 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2947 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2948 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2949 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2952 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2953 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2954 Intrinsic V4F32Int> {
2955 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2956 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2957 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2958 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2959 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2960 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2963 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2964 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2965 SDNode OpNode, Intrinsic F64Int> {
2966 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2967 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2968 [(set FR64:$dst, (OpNode FR64:$src))]>;
2969 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2970 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2971 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2972 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2973 Requires<[HasSSE2, OptForSize]>;
2974 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2975 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2976 [(set VR128:$dst, (F64Int VR128:$src))]>;
2977 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2978 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2979 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2982 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2983 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2984 let neverHasSideEffects = 1 in {
2985 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2986 !strconcat(OpcodeStr,
2987 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2989 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
2990 !strconcat(OpcodeStr,
2991 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2993 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
2994 (ins VR128:$src1, sdmem:$src2),
2995 !strconcat(OpcodeStr,
2996 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2999 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3000 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3002 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3003 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3004 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3005 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3006 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3007 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3010 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3011 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3012 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3013 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3014 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3015 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3016 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3017 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3020 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3021 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3022 Intrinsic V2F64Int> {
3023 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3024 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3025 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3026 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3027 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3028 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3031 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3032 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3033 Intrinsic V2F64Int> {
3034 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3035 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3036 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3037 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3038 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3039 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3042 let Predicates = [HasAVX] in {
3044 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3045 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3047 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3048 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3049 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3050 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3051 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3052 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3053 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3054 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3057 // Reciprocal approximations. Note that these typically require refinement
3058 // in order to obtain suitable precision.
3059 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3060 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3061 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3062 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3063 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3065 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3066 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3067 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3068 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3069 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3072 def : Pat<(f32 (fsqrt FR32:$src)),
3073 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3074 def : Pat<(f32 (fsqrt (load addr:$src))),
3075 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3076 Requires<[HasAVX, OptForSize]>;
3077 def : Pat<(f64 (fsqrt FR64:$src)),
3078 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3079 def : Pat<(f64 (fsqrt (load addr:$src))),
3080 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3081 Requires<[HasAVX, OptForSize]>;
3083 def : Pat<(f32 (X86frsqrt FR32:$src)),
3084 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3085 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3086 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3087 Requires<[HasAVX, OptForSize]>;
3089 def : Pat<(f32 (X86frcp FR32:$src)),
3090 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3091 def : Pat<(f32 (X86frcp (load addr:$src))),
3092 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3093 Requires<[HasAVX, OptForSize]>;
3095 let Predicates = [HasAVX] in {
3096 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3097 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3098 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3099 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3101 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3102 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3104 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3105 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3106 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3107 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3109 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3110 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3112 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3113 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3114 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3115 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3117 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3118 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3120 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3121 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3122 (VRCPSSr (f32 (IMPLICIT_DEF)),
3123 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3125 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3126 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3130 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3131 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3132 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3133 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3134 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3135 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3137 // Reciprocal approximations. Note that these typically require refinement
3138 // in order to obtain suitable precision.
3139 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3140 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3141 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3142 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3143 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3144 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3146 // There is no f64 version of the reciprocal approximation instructions.
3148 //===----------------------------------------------------------------------===//
3149 // SSE 1 & 2 - Non-temporal stores
3150 //===----------------------------------------------------------------------===//
3152 let AddedComplexity = 400 in { // Prefer non-temporal versions
3153 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3154 (ins f128mem:$dst, VR128:$src),
3155 "movntps\t{$src, $dst|$dst, $src}",
3156 [(alignednontemporalstore (v4f32 VR128:$src),
3158 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3159 (ins f128mem:$dst, VR128:$src),
3160 "movntpd\t{$src, $dst|$dst, $src}",
3161 [(alignednontemporalstore (v2f64 VR128:$src),
3163 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3164 (ins f128mem:$dst, VR128:$src),
3165 "movntdq\t{$src, $dst|$dst, $src}",
3166 [(alignednontemporalstore (v2f64 VR128:$src),
3169 let ExeDomain = SSEPackedInt in
3170 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3171 (ins f128mem:$dst, VR128:$src),
3172 "movntdq\t{$src, $dst|$dst, $src}",
3173 [(alignednontemporalstore (v4f32 VR128:$src),
3176 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3177 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3179 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3180 (ins f256mem:$dst, VR256:$src),
3181 "movntps\t{$src, $dst|$dst, $src}",
3182 [(alignednontemporalstore (v8f32 VR256:$src),
3184 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3185 (ins f256mem:$dst, VR256:$src),
3186 "movntpd\t{$src, $dst|$dst, $src}",
3187 [(alignednontemporalstore (v4f64 VR256:$src),
3189 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3190 (ins f256mem:$dst, VR256:$src),
3191 "movntdq\t{$src, $dst|$dst, $src}",
3192 [(alignednontemporalstore (v4f64 VR256:$src),
3194 let ExeDomain = SSEPackedInt in
3195 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3196 (ins f256mem:$dst, VR256:$src),
3197 "movntdq\t{$src, $dst|$dst, $src}",
3198 [(alignednontemporalstore (v8f32 VR256:$src),
3202 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3203 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3204 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3205 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3206 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3207 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3209 let AddedComplexity = 400 in { // Prefer non-temporal versions
3210 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3211 "movntps\t{$src, $dst|$dst, $src}",
3212 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3213 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3214 "movntpd\t{$src, $dst|$dst, $src}",
3215 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3217 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3218 "movntdq\t{$src, $dst|$dst, $src}",
3219 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3221 let ExeDomain = SSEPackedInt in
3222 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3223 "movntdq\t{$src, $dst|$dst, $src}",
3224 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3226 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3227 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3229 // There is no AVX form for instructions below this point
3230 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3231 "movnti{l}\t{$src, $dst|$dst, $src}",
3232 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3233 TB, Requires<[HasSSE2]>;
3234 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3235 "movnti{q}\t{$src, $dst|$dst, $src}",
3236 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3237 TB, Requires<[HasSSE2]>;
3240 //===----------------------------------------------------------------------===//
3241 // SSE 1 & 2 - Prefetch and memory fence
3242 //===----------------------------------------------------------------------===//
3244 // Prefetch intrinsic.
3245 def PREFETCHT0 : VoPSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3246 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3247 def PREFETCHT1 : VoPSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3248 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3249 def PREFETCHT2 : VoPSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3250 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3251 def PREFETCHNTA : VoPSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3252 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3255 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3256 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3257 TB, Requires<[HasSSE2]>;
3259 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3260 // was introduced with SSE2, it's backward compatible.
3261 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3263 // Load, store, and memory fence
3264 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3265 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3266 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3267 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3268 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3269 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3271 def : Pat<(X86SFence), (SFENCE)>;
3272 def : Pat<(X86LFence), (LFENCE)>;
3273 def : Pat<(X86MFence), (MFENCE)>;
3275 //===----------------------------------------------------------------------===//
3276 // SSE 1 & 2 - Load/Store XCSR register
3277 //===----------------------------------------------------------------------===//
3279 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3280 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3281 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3282 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3284 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3285 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3286 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3287 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3289 //===---------------------------------------------------------------------===//
3290 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3291 //===---------------------------------------------------------------------===//
3293 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3295 let neverHasSideEffects = 1 in {
3296 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3297 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3298 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3299 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3301 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3302 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3303 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3304 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3307 let isCodeGenOnly = 1 in {
3308 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3309 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3310 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3311 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3312 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3313 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3314 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3315 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3318 let canFoldAsLoad = 1, mayLoad = 1 in {
3319 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3320 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3321 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3322 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3323 let Predicates = [HasAVX] in {
3324 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3325 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3326 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3327 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3331 let mayStore = 1 in {
3332 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3333 (ins i128mem:$dst, VR128:$src),
3334 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3335 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3336 (ins i256mem:$dst, VR256:$src),
3337 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3338 let Predicates = [HasAVX] in {
3339 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3340 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3341 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3342 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3346 let neverHasSideEffects = 1 in
3347 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3348 "movdqa\t{$src, $dst|$dst, $src}", []>;
3350 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3351 "movdqu\t{$src, $dst|$dst, $src}",
3352 []>, XS, Requires<[HasSSE2]>;
3355 let isCodeGenOnly = 1 in {
3356 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3357 "movdqa\t{$src, $dst|$dst, $src}", []>;
3359 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3360 "movdqu\t{$src, $dst|$dst, $src}",
3361 []>, XS, Requires<[HasSSE2]>;
3364 let canFoldAsLoad = 1, mayLoad = 1 in {
3365 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3366 "movdqa\t{$src, $dst|$dst, $src}",
3367 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3368 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3369 "movdqu\t{$src, $dst|$dst, $src}",
3370 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3371 XS, Requires<[HasSSE2]>;
3374 let mayStore = 1 in {
3375 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3376 "movdqa\t{$src, $dst|$dst, $src}",
3377 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3378 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3379 "movdqu\t{$src, $dst|$dst, $src}",
3380 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3381 XS, Requires<[HasSSE2]>;
3384 // Intrinsic forms of MOVDQU load and store
3385 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3386 "vmovdqu\t{$src, $dst|$dst, $src}",
3387 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3388 XS, VEX, Requires<[HasAVX]>;
3390 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3391 "movdqu\t{$src, $dst|$dst, $src}",
3392 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3393 XS, Requires<[HasSSE2]>;
3395 } // ExeDomain = SSEPackedInt
3397 let Predicates = [HasAVX] in {
3398 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3399 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3400 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3403 //===---------------------------------------------------------------------===//
3404 // SSE2 - Packed Integer Arithmetic Instructions
3405 //===---------------------------------------------------------------------===//
3407 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3409 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3410 RegisterClass RC, PatFrag memop_frag,
3411 X86MemOperand x86memop, bit IsCommutable = 0,
3413 let isCommutable = IsCommutable in
3414 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3415 (ins RC:$src1, RC:$src2),
3417 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3418 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3419 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3420 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3421 (ins RC:$src1, x86memop:$src2),
3423 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3424 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3425 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3428 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3429 string OpcodeStr, Intrinsic IntId,
3430 Intrinsic IntId2, RegisterClass RC,
3432 // src2 is always 128-bit
3433 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3434 (ins RC:$src1, VR128:$src2),
3436 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3437 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3438 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3439 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3440 (ins RC:$src1, i128mem:$src2),
3442 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3443 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3444 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3445 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3446 (ins RC:$src1, i32i8imm:$src2),
3448 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3449 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3450 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3453 /// PDI_binop_rm - Simple SSE2 binary operator.
3454 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3455 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
3456 X86MemOperand x86memop, bit IsCommutable = 0,
3458 let isCommutable = IsCommutable in
3459 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3460 (ins RC:$src1, RC:$src2),
3462 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3463 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3464 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
3465 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3466 (ins RC:$src1, x86memop:$src2),
3468 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3469 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3470 [(set RC:$dst, (OpVT (OpNode RC:$src1,
3471 (bitconvert (memop_frag addr:$src2)))))]>;
3473 } // ExeDomain = SSEPackedInt
3475 // 128-bit Integer Arithmetic
3477 let Predicates = [HasAVX] in {
3478 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3479 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3480 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3481 i128mem, 1, 0>, VEX_4V;
3482 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3483 i128mem, 1, 0>, VEX_4V;
3484 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3485 i128mem, 1, 0>, VEX_4V;
3486 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3487 i128mem, 1, 0>, VEX_4V;
3488 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3489 i128mem, 0, 0>, VEX_4V;
3490 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3491 i128mem, 0, 0>, VEX_4V;
3492 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3493 i128mem, 0, 0>, VEX_4V;
3494 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3495 i128mem, 0, 0>, VEX_4V;
3498 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3499 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3500 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3501 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3502 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3503 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3504 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3505 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3506 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3507 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3508 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3509 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3510 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3511 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3512 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3513 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3514 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3515 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3516 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3517 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3518 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3519 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3520 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3521 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3522 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3523 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3524 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3525 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3526 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3527 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3528 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3529 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3530 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3531 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3532 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3533 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3534 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3535 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3538 let Predicates = [HasAVX2] in {
3539 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3540 i256mem, 1, 0>, VEX_4V;
3541 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3542 i256mem, 1, 0>, VEX_4V;
3543 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3544 i256mem, 1, 0>, VEX_4V;
3545 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3546 i256mem, 1, 0>, VEX_4V;
3547 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3548 i256mem, 1, 0>, VEX_4V;
3549 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3550 i256mem, 0, 0>, VEX_4V;
3551 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3552 i256mem, 0, 0>, VEX_4V;
3553 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3554 i256mem, 0, 0>, VEX_4V;
3555 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3556 i256mem, 0, 0>, VEX_4V;
3559 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3560 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3561 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3562 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3563 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3564 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3565 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3566 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3567 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3568 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3569 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3570 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3571 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3572 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3573 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3574 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3575 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3576 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3577 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3578 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3579 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3580 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3581 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3582 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3583 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3584 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3585 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3586 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3587 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3588 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3589 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3590 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3591 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3592 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3593 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3594 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3595 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3596 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3599 let Constraints = "$src1 = $dst" in {
3600 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3602 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3604 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3606 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3608 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3610 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3612 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3614 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3616 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3620 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3621 VR128, memopv2i64, i128mem>;
3622 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3623 VR128, memopv2i64, i128mem>;
3624 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3625 VR128, memopv2i64, i128mem>;
3626 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3627 VR128, memopv2i64, i128mem>;
3628 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3629 VR128, memopv2i64, i128mem, 1>;
3630 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3631 VR128, memopv2i64, i128mem, 1>;
3632 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3633 VR128, memopv2i64, i128mem, 1>;
3634 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3635 VR128, memopv2i64, i128mem, 1>;
3636 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3637 VR128, memopv2i64, i128mem, 1>;
3638 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3639 VR128, memopv2i64, i128mem, 1>;
3640 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3641 VR128, memopv2i64, i128mem, 1>;
3642 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3643 VR128, memopv2i64, i128mem, 1>;
3644 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3645 VR128, memopv2i64, i128mem, 1>;
3646 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3647 VR128, memopv2i64, i128mem, 1>;
3648 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3649 VR128, memopv2i64, i128mem, 1>;
3650 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3651 VR128, memopv2i64, i128mem, 1>;
3652 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3653 VR128, memopv2i64, i128mem, 1>;
3654 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3655 VR128, memopv2i64, i128mem, 1>;
3656 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3657 VR128, memopv2i64, i128mem, 1>;
3659 } // Constraints = "$src1 = $dst"
3661 //===---------------------------------------------------------------------===//
3662 // SSE2 - Packed Integer Logical Instructions
3663 //===---------------------------------------------------------------------===//
3665 let Predicates = [HasAVX] in {
3666 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3667 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3669 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3670 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3672 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3673 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3676 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3677 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3679 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3680 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3682 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3683 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3686 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3687 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3689 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3690 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3693 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
3694 i128mem, 1, 0>, VEX_4V;
3695 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
3696 i128mem, 1, 0>, VEX_4V;
3697 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
3698 i128mem, 1, 0>, VEX_4V;
3699 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
3700 i128mem, 0, 0>, VEX_4V;
3702 let ExeDomain = SSEPackedInt in {
3703 let neverHasSideEffects = 1 in {
3704 // 128-bit logical shifts.
3705 def VPSLLDQri : PDIi8<0x73, MRM7r,
3706 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3707 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3709 def VPSRLDQri : PDIi8<0x73, MRM3r,
3710 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3711 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3713 // PSRADQri doesn't exist in SSE[1-3].
3718 let Predicates = [HasAVX2] in {
3719 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3720 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3722 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3723 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3725 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3726 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3729 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3730 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3732 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3733 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3735 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3736 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3739 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3740 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3742 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3743 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3746 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
3747 i256mem, 1, 0>, VEX_4V;
3748 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
3749 i256mem, 1, 0>, VEX_4V;
3750 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
3751 i256mem, 1, 0>, VEX_4V;
3752 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
3753 i256mem, 0, 0>, VEX_4V;
3755 let ExeDomain = SSEPackedInt in {
3756 let neverHasSideEffects = 1 in {
3757 // 128-bit logical shifts.
3758 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3759 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3760 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3762 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3763 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3764 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3766 // PSRADQYri doesn't exist in SSE[1-3].
3771 let Constraints = "$src1 = $dst" in {
3772 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3773 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3775 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3776 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3778 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3779 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3782 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3783 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3785 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3786 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3788 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3789 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3792 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3793 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3795 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3796 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3799 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
3801 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
3803 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
3805 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
3808 let ExeDomain = SSEPackedInt in {
3809 let neverHasSideEffects = 1 in {
3810 // 128-bit logical shifts.
3811 def PSLLDQri : PDIi8<0x73, MRM7r,
3812 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3813 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3814 def PSRLDQri : PDIi8<0x73, MRM3r,
3815 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3816 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3817 // PSRADQri doesn't exist in SSE[1-3].
3820 } // Constraints = "$src1 = $dst"
3822 let Predicates = [HasAVX] in {
3823 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3824 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3825 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3826 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3827 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3828 (VPSLLDQri VR128:$src1, imm:$src2)>;
3829 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3830 (VPSRLDQri VR128:$src1, imm:$src2)>;
3831 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3832 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3834 // Shift up / down and insert zero's.
3835 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3836 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3837 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3838 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3841 let Predicates = [HasAVX2] in {
3842 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3843 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3844 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3845 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3846 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3847 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3848 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3849 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3852 let Predicates = [HasSSE2] in {
3853 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3854 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3855 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3856 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3857 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3858 (PSLLDQri VR128:$src1, imm:$src2)>;
3859 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3860 (PSRLDQri VR128:$src1, imm:$src2)>;
3861 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3862 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3864 // Shift up / down and insert zero's.
3865 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3866 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3867 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3868 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3871 //===---------------------------------------------------------------------===//
3872 // SSE2 - Packed Integer Comparison Instructions
3873 //===---------------------------------------------------------------------===//
3875 let Predicates = [HasAVX] in {
3876 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3877 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3878 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3879 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3880 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3881 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3882 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3883 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3884 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3885 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3886 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3887 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3889 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3890 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3891 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3892 (bc_v16i8 (memopv2i64 addr:$src2)))),
3893 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3894 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3895 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3896 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3897 (bc_v8i16 (memopv2i64 addr:$src2)))),
3898 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3899 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3900 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3901 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3902 (bc_v4i32 (memopv2i64 addr:$src2)))),
3903 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3905 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3906 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3907 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3908 (bc_v16i8 (memopv2i64 addr:$src2)))),
3909 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3910 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3911 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3912 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3913 (bc_v8i16 (memopv2i64 addr:$src2)))),
3914 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3915 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3916 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3917 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3918 (bc_v4i32 (memopv2i64 addr:$src2)))),
3919 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3922 let Predicates = [HasAVX2] in {
3923 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3924 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3925 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3926 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3927 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3928 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3929 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3930 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3931 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3932 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3933 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3934 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3936 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3937 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3938 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3939 (bc_v32i8 (memopv4i64 addr:$src2)))),
3940 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3941 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3942 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3943 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3944 (bc_v16i16 (memopv4i64 addr:$src2)))),
3945 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3946 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3947 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3948 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3949 (bc_v8i32 (memopv4i64 addr:$src2)))),
3950 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3952 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3953 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3954 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3955 (bc_v32i8 (memopv4i64 addr:$src2)))),
3956 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3957 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3958 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3959 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3960 (bc_v16i16 (memopv4i64 addr:$src2)))),
3961 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3962 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3963 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3964 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3965 (bc_v8i32 (memopv4i64 addr:$src2)))),
3966 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3969 let Constraints = "$src1 = $dst" in {
3970 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3971 VR128, memopv2i64, i128mem, 1>;
3972 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3973 VR128, memopv2i64, i128mem, 1>;
3974 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3975 VR128, memopv2i64, i128mem, 1>;
3976 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3977 VR128, memopv2i64, i128mem>;
3978 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3979 VR128, memopv2i64, i128mem>;
3980 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3981 VR128, memopv2i64, i128mem>;
3982 } // Constraints = "$src1 = $dst"
3984 let Predicates = [HasSSE2] in {
3985 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3986 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3987 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3988 (bc_v16i8 (memopv2i64 addr:$src2)))),
3989 (PCMPEQBrm VR128:$src1, addr:$src2)>;
3990 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3991 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
3992 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3993 (bc_v8i16 (memopv2i64 addr:$src2)))),
3994 (PCMPEQWrm VR128:$src1, addr:$src2)>;
3995 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3996 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
3997 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3998 (bc_v4i32 (memopv2i64 addr:$src2)))),
3999 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4001 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4002 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4003 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4004 (bc_v16i8 (memopv2i64 addr:$src2)))),
4005 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4006 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4007 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4008 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4009 (bc_v8i16 (memopv2i64 addr:$src2)))),
4010 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4011 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4012 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4013 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4014 (bc_v4i32 (memopv2i64 addr:$src2)))),
4015 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4018 //===---------------------------------------------------------------------===//
4019 // SSE2 - Packed Integer Pack Instructions
4020 //===---------------------------------------------------------------------===//
4022 let Predicates = [HasAVX] in {
4023 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4024 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4025 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4026 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4027 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4028 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4031 let Predicates = [HasAVX2] in {
4032 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4033 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4034 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4035 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4036 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4037 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4040 let Constraints = "$src1 = $dst" in {
4041 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4042 VR128, memopv2i64, i128mem>;
4043 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4044 VR128, memopv2i64, i128mem>;
4045 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4046 VR128, memopv2i64, i128mem>;
4047 } // Constraints = "$src1 = $dst"
4049 //===---------------------------------------------------------------------===//
4050 // SSE2 - Packed Integer Shuffle Instructions
4051 //===---------------------------------------------------------------------===//
4053 let ExeDomain = SSEPackedInt in {
4054 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4056 def ri : Ii8<0x70, MRMSrcReg,
4057 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4058 !strconcat(OpcodeStr,
4059 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4060 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4062 def mi : Ii8<0x70, MRMSrcMem,
4063 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4064 !strconcat(OpcodeStr,
4065 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4066 [(set VR128:$dst, (vt (pshuf_frag:$src2
4067 (bc_frag (memopv2i64 addr:$src1)),
4071 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4073 def Yri : Ii8<0x70, MRMSrcReg,
4074 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4075 !strconcat(OpcodeStr,
4076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4077 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4079 def Ymi : Ii8<0x70, MRMSrcMem,
4080 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4081 !strconcat(OpcodeStr,
4082 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4083 [(set VR256:$dst, (vt (pshuf_frag:$src2
4084 (bc_frag (memopv4i64 addr:$src1)),
4087 } // ExeDomain = SSEPackedInt
4089 let Predicates = [HasAVX] in {
4090 let AddedComplexity = 5 in
4091 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4094 // SSE2 with ImmT == Imm8 and XS prefix.
4095 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4098 // SSE2 with ImmT == Imm8 and XD prefix.
4099 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4102 let AddedComplexity = 5 in
4103 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4104 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4105 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4106 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4107 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4109 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4111 (VPSHUFDmi addr:$src1, imm:$imm)>;
4112 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4114 (VPSHUFDmi addr:$src1, imm:$imm)>;
4115 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4116 (VPSHUFDri VR128:$src1, imm:$imm)>;
4117 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4118 (VPSHUFDri VR128:$src1, imm:$imm)>;
4119 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4120 (VPSHUFHWri VR128:$src, imm:$imm)>;
4121 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4123 (VPSHUFHWmi addr:$src, imm:$imm)>;
4124 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4125 (VPSHUFLWri VR128:$src, imm:$imm)>;
4126 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4128 (VPSHUFLWmi addr:$src, imm:$imm)>;
4131 let Predicates = [HasAVX2] in {
4132 let AddedComplexity = 5 in
4133 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4136 // SSE2 with ImmT == Imm8 and XS prefix.
4137 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4140 // SSE2 with ImmT == Imm8 and XD prefix.
4141 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4145 let Predicates = [HasSSE2] in {
4146 let AddedComplexity = 5 in
4147 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4149 // SSE2 with ImmT == Imm8 and XS prefix.
4150 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4152 // SSE2 with ImmT == Imm8 and XD prefix.
4153 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4155 let AddedComplexity = 5 in
4156 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4157 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4158 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4159 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4160 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4162 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4164 (PSHUFDmi addr:$src1, imm:$imm)>;
4165 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4167 (PSHUFDmi addr:$src1, imm:$imm)>;
4168 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4169 (PSHUFDri VR128:$src1, imm:$imm)>;
4170 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4171 (PSHUFDri VR128:$src1, imm:$imm)>;
4172 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4173 (PSHUFHWri VR128:$src, imm:$imm)>;
4174 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4176 (PSHUFHWmi addr:$src, imm:$imm)>;
4177 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4178 (PSHUFLWri VR128:$src, imm:$imm)>;
4179 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4181 (PSHUFLWmi addr:$src, imm:$imm)>;
4184 //===---------------------------------------------------------------------===//
4185 // SSE2 - Packed Integer Unpack Instructions
4186 //===---------------------------------------------------------------------===//
4188 let ExeDomain = SSEPackedInt in {
4189 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4190 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4191 def rr : PDI<opc, MRMSrcReg,
4192 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4194 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4195 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4196 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4197 def rm : PDI<opc, MRMSrcMem,
4198 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4200 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4201 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4202 [(set VR128:$dst, (OpNode VR128:$src1,
4203 (bc_frag (memopv2i64
4207 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4208 SDNode OpNode, PatFrag bc_frag> {
4209 def Yrr : PDI<opc, MRMSrcReg,
4210 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4211 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4212 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4213 def Yrm : PDI<opc, MRMSrcMem,
4214 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4215 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4216 [(set VR256:$dst, (OpNode VR256:$src1,
4217 (bc_frag (memopv4i64 addr:$src2))))]>;
4220 let Predicates = [HasAVX] in {
4221 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4222 bc_v16i8, 0>, VEX_4V;
4223 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4224 bc_v8i16, 0>, VEX_4V;
4225 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4226 bc_v4i32, 0>, VEX_4V;
4227 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4228 bc_v2i64, 0>, VEX_4V;
4230 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4231 bc_v16i8, 0>, VEX_4V;
4232 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4233 bc_v8i16, 0>, VEX_4V;
4234 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4235 bc_v4i32, 0>, VEX_4V;
4236 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4237 bc_v2i64, 0>, VEX_4V;
4240 let Predicates = [HasAVX2] in {
4241 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4243 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4245 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4247 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4250 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4252 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4254 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4256 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4260 let Constraints = "$src1 = $dst" in {
4261 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4263 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4265 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4267 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4270 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4272 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4274 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4276 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4279 } // ExeDomain = SSEPackedInt
4281 // Patterns for using AVX1 instructions with integer vectors
4282 // Here to give AVX2 priority
4283 let Predicates = [HasAVX] in {
4284 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4285 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4286 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4287 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4288 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4289 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4290 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4291 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4293 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4294 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4295 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4296 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4297 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4298 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4299 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4300 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4303 // Splat v2f64 / v2i64
4304 let AddedComplexity = 10 in {
4305 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4306 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4307 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4308 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4311 //===---------------------------------------------------------------------===//
4312 // SSE2 - Packed Integer Extract and Insert
4313 //===---------------------------------------------------------------------===//
4315 let ExeDomain = SSEPackedInt in {
4316 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4317 def rri : Ii8<0xC4, MRMSrcReg,
4318 (outs VR128:$dst), (ins VR128:$src1,
4319 GR32:$src2, i32i8imm:$src3),
4321 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4322 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4324 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4325 def rmi : Ii8<0xC4, MRMSrcMem,
4326 (outs VR128:$dst), (ins VR128:$src1,
4327 i16mem:$src2, i32i8imm:$src3),
4329 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4330 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4332 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4337 let Predicates = [HasAVX] in
4338 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4339 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4340 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4341 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4342 imm:$src2))]>, TB, OpSize, VEX;
4343 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4344 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4345 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4346 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4350 let Predicates = [HasAVX] in {
4351 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4352 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4353 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4354 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4355 []>, TB, OpSize, VEX_4V;
4358 let Constraints = "$src1 = $dst" in
4359 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4361 } // ExeDomain = SSEPackedInt
4363 //===---------------------------------------------------------------------===//
4364 // SSE2 - Packed Mask Creation
4365 //===---------------------------------------------------------------------===//
4367 let ExeDomain = SSEPackedInt in {
4369 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4370 "pmovmskb\t{$src, $dst|$dst, $src}",
4371 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4372 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4373 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4375 let Predicates = [HasAVX2] in {
4376 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4377 "pmovmskb\t{$src, $dst|$dst, $src}",
4378 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4379 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4380 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4383 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4384 "pmovmskb\t{$src, $dst|$dst, $src}",
4385 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4387 } // ExeDomain = SSEPackedInt
4389 //===---------------------------------------------------------------------===//
4390 // SSE2 - Conditional Store
4391 //===---------------------------------------------------------------------===//
4393 let ExeDomain = SSEPackedInt in {
4396 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4397 (ins VR128:$src, VR128:$mask),
4398 "maskmovdqu\t{$mask, $src|$src, $mask}",
4399 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4401 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4402 (ins VR128:$src, VR128:$mask),
4403 "maskmovdqu\t{$mask, $src|$src, $mask}",
4404 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4407 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4408 "maskmovdqu\t{$mask, $src|$src, $mask}",
4409 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4411 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4412 "maskmovdqu\t{$mask, $src|$src, $mask}",
4413 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4415 } // ExeDomain = SSEPackedInt
4417 //===---------------------------------------------------------------------===//
4418 // SSE2 - Move Doubleword
4419 //===---------------------------------------------------------------------===//
4421 //===---------------------------------------------------------------------===//
4422 // Move Int Doubleword to Packed Double Int
4424 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4425 "movd\t{$src, $dst|$dst, $src}",
4427 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4428 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4429 "movd\t{$src, $dst|$dst, $src}",
4431 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4433 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4434 "mov{d|q}\t{$src, $dst|$dst, $src}",
4436 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4437 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4438 "mov{d|q}\t{$src, $dst|$dst, $src}",
4439 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4441 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4442 "movd\t{$src, $dst|$dst, $src}",
4444 (v4i32 (scalar_to_vector GR32:$src)))]>;
4445 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4446 "movd\t{$src, $dst|$dst, $src}",
4448 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4449 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4450 "mov{d|q}\t{$src, $dst|$dst, $src}",
4452 (v2i64 (scalar_to_vector GR64:$src)))]>;
4453 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4454 "mov{d|q}\t{$src, $dst|$dst, $src}",
4455 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4457 //===---------------------------------------------------------------------===//
4458 // Move Int Doubleword to Single Scalar
4460 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4461 "movd\t{$src, $dst|$dst, $src}",
4462 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4464 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4465 "movd\t{$src, $dst|$dst, $src}",
4466 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4468 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4469 "movd\t{$src, $dst|$dst, $src}",
4470 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4472 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4473 "movd\t{$src, $dst|$dst, $src}",
4474 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4476 //===---------------------------------------------------------------------===//
4477 // Move Packed Doubleword Int to Packed Double Int
4479 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4480 "movd\t{$src, $dst|$dst, $src}",
4481 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4483 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4484 (ins i32mem:$dst, VR128:$src),
4485 "movd\t{$src, $dst|$dst, $src}",
4486 [(store (i32 (vector_extract (v4i32 VR128:$src),
4487 (iPTR 0))), addr:$dst)]>, VEX;
4488 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4489 "movd\t{$src, $dst|$dst, $src}",
4490 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4492 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4493 "movd\t{$src, $dst|$dst, $src}",
4494 [(store (i32 (vector_extract (v4i32 VR128:$src),
4495 (iPTR 0))), addr:$dst)]>;
4497 //===---------------------------------------------------------------------===//
4498 // Move Packed Doubleword Int first element to Doubleword Int
4500 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4501 "mov{d|q}\t{$src, $dst|$dst, $src}",
4502 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4504 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4506 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4507 "mov{d|q}\t{$src, $dst|$dst, $src}",
4508 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4511 //===---------------------------------------------------------------------===//
4512 // Bitcast FR64 <-> GR64
4514 let Predicates = [HasAVX] in
4515 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4516 "vmovq\t{$src, $dst|$dst, $src}",
4517 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4519 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4520 "mov{d|q}\t{$src, $dst|$dst, $src}",
4521 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4522 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4523 "movq\t{$src, $dst|$dst, $src}",
4524 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4526 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4527 "movq\t{$src, $dst|$dst, $src}",
4528 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4529 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4530 "mov{d|q}\t{$src, $dst|$dst, $src}",
4531 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4532 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4533 "movq\t{$src, $dst|$dst, $src}",
4534 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4536 //===---------------------------------------------------------------------===//
4537 // Move Scalar Single to Double Int
4539 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4540 "movd\t{$src, $dst|$dst, $src}",
4541 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4542 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4543 "movd\t{$src, $dst|$dst, $src}",
4544 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4545 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4546 "movd\t{$src, $dst|$dst, $src}",
4547 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4548 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4549 "movd\t{$src, $dst|$dst, $src}",
4550 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4552 //===---------------------------------------------------------------------===//
4553 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4555 let AddedComplexity = 15 in {
4556 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4557 "movd\t{$src, $dst|$dst, $src}",
4558 [(set VR128:$dst, (v4i32 (X86vzmovl
4559 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4561 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4562 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4563 [(set VR128:$dst, (v2i64 (X86vzmovl
4564 (v2i64 (scalar_to_vector GR64:$src)))))]>,
4567 let AddedComplexity = 15 in {
4568 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4569 "movd\t{$src, $dst|$dst, $src}",
4570 [(set VR128:$dst, (v4i32 (X86vzmovl
4571 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4572 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4573 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4574 [(set VR128:$dst, (v2i64 (X86vzmovl
4575 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4578 let AddedComplexity = 20 in {
4579 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4580 "movd\t{$src, $dst|$dst, $src}",
4582 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4583 (loadi32 addr:$src))))))]>,
4585 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4586 "movd\t{$src, $dst|$dst, $src}",
4588 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4589 (loadi32 addr:$src))))))]>;
4592 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4593 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4594 (MOVZDI2PDIrm addr:$src)>;
4595 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4596 (MOVZDI2PDIrm addr:$src)>;
4597 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4598 (MOVZDI2PDIrm addr:$src)>;
4601 let Predicates = [HasAVX] in {
4602 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4603 let AddedComplexity = 20 in {
4604 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4605 (VMOVZDI2PDIrm addr:$src)>;
4606 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4607 (VMOVZDI2PDIrm addr:$src)>;
4608 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4609 (VMOVZDI2PDIrm addr:$src)>;
4611 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4612 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4613 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4614 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4615 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4616 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4617 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4620 // These are the correct encodings of the instructions so that we know how to
4621 // read correct assembly, even though we continue to emit the wrong ones for
4622 // compatibility with Darwin's buggy assembler.
4623 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4624 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4625 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4626 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4627 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4628 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4629 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4630 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4631 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4632 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4633 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4634 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4636 //===---------------------------------------------------------------------===//
4637 // SSE2 - Move Quadword
4638 //===---------------------------------------------------------------------===//
4640 //===---------------------------------------------------------------------===//
4641 // Move Quadword Int to Packed Quadword Int
4643 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4644 "vmovq\t{$src, $dst|$dst, $src}",
4646 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4647 VEX, Requires<[HasAVX]>;
4648 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4649 "movq\t{$src, $dst|$dst, $src}",
4651 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4652 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4654 //===---------------------------------------------------------------------===//
4655 // Move Packed Quadword Int to Quadword Int
4657 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4658 "movq\t{$src, $dst|$dst, $src}",
4659 [(store (i64 (vector_extract (v2i64 VR128:$src),
4660 (iPTR 0))), addr:$dst)]>, VEX;
4661 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4662 "movq\t{$src, $dst|$dst, $src}",
4663 [(store (i64 (vector_extract (v2i64 VR128:$src),
4664 (iPTR 0))), addr:$dst)]>;
4666 //===---------------------------------------------------------------------===//
4667 // Store / copy lower 64-bits of a XMM register.
4669 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4670 "movq\t{$src, $dst|$dst, $src}",
4671 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4672 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4673 "movq\t{$src, $dst|$dst, $src}",
4674 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4676 let AddedComplexity = 20 in
4677 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4678 "vmovq\t{$src, $dst|$dst, $src}",
4680 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4681 (loadi64 addr:$src))))))]>,
4682 XS, VEX, Requires<[HasAVX]>;
4684 let AddedComplexity = 20 in
4685 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4686 "movq\t{$src, $dst|$dst, $src}",
4688 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4689 (loadi64 addr:$src))))))]>,
4690 XS, Requires<[HasSSE2]>;
4692 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4693 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4694 (MOVZQI2PQIrm addr:$src)>;
4695 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4696 (MOVZQI2PQIrm addr:$src)>;
4697 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4700 let Predicates = [HasAVX], AddedComplexity = 20 in {
4701 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4702 (VMOVZQI2PQIrm addr:$src)>;
4703 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4704 (VMOVZQI2PQIrm addr:$src)>;
4705 def : Pat<(v2i64 (X86vzload addr:$src)),
4706 (VMOVZQI2PQIrm addr:$src)>;
4709 //===---------------------------------------------------------------------===//
4710 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4711 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4713 let AddedComplexity = 15 in
4714 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4715 "vmovq\t{$src, $dst|$dst, $src}",
4716 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4717 XS, VEX, Requires<[HasAVX]>;
4718 let AddedComplexity = 15 in
4719 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4720 "movq\t{$src, $dst|$dst, $src}",
4721 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4722 XS, Requires<[HasSSE2]>;
4724 let AddedComplexity = 20 in
4725 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4726 "vmovq\t{$src, $dst|$dst, $src}",
4727 [(set VR128:$dst, (v2i64 (X86vzmovl
4728 (loadv2i64 addr:$src))))]>,
4729 XS, VEX, Requires<[HasAVX]>;
4730 let AddedComplexity = 20 in {
4731 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4732 "movq\t{$src, $dst|$dst, $src}",
4733 [(set VR128:$dst, (v2i64 (X86vzmovl
4734 (loadv2i64 addr:$src))))]>,
4735 XS, Requires<[HasSSE2]>;
4738 let AddedComplexity = 20 in {
4739 let Predicates = [HasSSE2] in {
4740 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4741 (MOVZPQILo2PQIrm addr:$src)>;
4742 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4743 (MOVZPQILo2PQIrr VR128:$src)>;
4745 let Predicates = [HasAVX] in {
4746 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4747 (VMOVZPQILo2PQIrm addr:$src)>;
4748 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4749 (VMOVZPQILo2PQIrr VR128:$src)>;
4753 // Instructions to match in the assembler
4754 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4755 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4756 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4757 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4758 // Recognize "movd" with GR64 destination, but encode as a "movq"
4759 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4760 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4762 // Instructions for the disassembler
4763 // xr = XMM register
4766 let Predicates = [HasAVX] in
4767 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4768 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4769 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4770 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4772 //===---------------------------------------------------------------------===//
4773 // SSE3 - Conversion Instructions
4774 //===---------------------------------------------------------------------===//
4776 // Convert Packed Double FP to Packed DW Integers
4777 let Predicates = [HasAVX] in {
4778 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4779 // register, but the same isn't true when using memory operands instead.
4780 // Provide other assembly rr and rm forms to address this explicitly.
4781 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4782 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4783 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4784 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4787 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4788 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4789 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4790 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4793 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4794 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4795 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4796 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4799 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4800 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4801 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4802 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4804 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4805 (VCVTPD2DQYrr VR256:$src)>;
4806 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4807 (VCVTPD2DQYrm addr:$src)>;
4809 // Convert Packed DW Integers to Packed Double FP
4810 let Predicates = [HasAVX] in {
4811 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4812 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4813 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4814 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4815 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4816 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4817 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4818 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4821 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4822 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4823 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4824 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4826 // AVX 256-bit register conversion intrinsics
4827 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4828 (VCVTDQ2PDYrr VR128:$src)>;
4829 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4830 (VCVTDQ2PDYrm addr:$src)>;
4832 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4833 (VCVTPD2DQYrr VR256:$src)>;
4834 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4835 (VCVTPD2DQYrm addr:$src)>;
4837 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4838 (VCVTDQ2PDYrr VR128:$src)>;
4839 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4840 (VCVTDQ2PDYrm addr:$src)>;
4842 //===---------------------------------------------------------------------===//
4843 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4844 //===---------------------------------------------------------------------===//
4845 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4846 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4847 X86MemOperand x86memop> {
4848 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4849 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4850 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4851 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4852 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4853 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4856 let Predicates = [HasAVX] in {
4857 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4858 v4f32, VR128, memopv4f32, f128mem>, VEX;
4859 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4860 v4f32, VR128, memopv4f32, f128mem>, VEX;
4861 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4862 v8f32, VR256, memopv8f32, f256mem>, VEX;
4863 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4864 v8f32, VR256, memopv8f32, f256mem>, VEX;
4866 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4867 memopv4f32, f128mem>;
4868 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4869 memopv4f32, f128mem>;
4871 let Predicates = [HasSSE3] in {
4872 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4873 (MOVSHDUPrr VR128:$src)>;
4874 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4875 (MOVSHDUPrm addr:$src)>;
4876 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4877 (MOVSLDUPrr VR128:$src)>;
4878 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4879 (MOVSLDUPrm addr:$src)>;
4882 let Predicates = [HasAVX] in {
4883 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4884 (VMOVSHDUPrr VR128:$src)>;
4885 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4886 (VMOVSHDUPrm addr:$src)>;
4887 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4888 (VMOVSLDUPrr VR128:$src)>;
4889 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4890 (VMOVSLDUPrm addr:$src)>;
4891 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4892 (VMOVSHDUPYrr VR256:$src)>;
4893 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4894 (VMOVSHDUPYrm addr:$src)>;
4895 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4896 (VMOVSLDUPYrr VR256:$src)>;
4897 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4898 (VMOVSLDUPYrm addr:$src)>;
4901 //===---------------------------------------------------------------------===//
4902 // SSE3 - Replicate Double FP - MOVDDUP
4903 //===---------------------------------------------------------------------===//
4905 multiclass sse3_replicate_dfp<string OpcodeStr> {
4906 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4907 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4908 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4909 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4910 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4912 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4916 // FIXME: Merge with above classe when there're patterns for the ymm version
4917 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4918 let Predicates = [HasAVX] in {
4919 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4920 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4922 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4923 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4928 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4929 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4930 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4932 let Predicates = [HasSSE3] in {
4933 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4935 (MOVDDUPrm addr:$src)>;
4936 let AddedComplexity = 5 in {
4937 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4938 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4939 (MOVDDUPrm addr:$src)>;
4940 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4941 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4942 (MOVDDUPrm addr:$src)>;
4944 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4945 (MOVDDUPrm addr:$src)>;
4946 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4947 (MOVDDUPrm addr:$src)>;
4948 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4949 (MOVDDUPrm addr:$src)>;
4950 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4951 (MOVDDUPrm addr:$src)>;
4952 def : Pat<(X86Movddup (bc_v2f64
4953 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4954 (MOVDDUPrm addr:$src)>;
4957 let Predicates = [HasAVX] in {
4958 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4960 (VMOVDDUPrm addr:$src)>;
4961 let AddedComplexity = 5 in {
4962 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4963 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4964 (VMOVDDUPrm addr:$src)>;
4965 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4966 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4967 (VMOVDDUPrm addr:$src)>;
4969 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4970 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4971 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4972 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4973 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4974 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4975 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4976 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4977 def : Pat<(X86Movddup (bc_v2f64
4978 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4979 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4982 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4983 (VMOVDDUPYrm addr:$src)>;
4984 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4985 (VMOVDDUPYrm addr:$src)>;
4986 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4987 (VMOVDDUPYrm addr:$src)>;
4988 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
4989 (VMOVDDUPYrm addr:$src)>;
4990 def : Pat<(X86Movddup (v4f64 VR256:$src)),
4991 (VMOVDDUPYrr VR256:$src)>;
4992 def : Pat<(X86Movddup (v4i64 VR256:$src)),
4993 (VMOVDDUPYrr VR256:$src)>;
4996 //===---------------------------------------------------------------------===//
4997 // SSE3 - Move Unaligned Integer
4998 //===---------------------------------------------------------------------===//
5000 let Predicates = [HasAVX] in {
5001 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5002 "vlddqu\t{$src, $dst|$dst, $src}",
5003 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5004 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5005 "vlddqu\t{$src, $dst|$dst, $src}",
5006 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5008 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5009 "lddqu\t{$src, $dst|$dst, $src}",
5010 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5012 //===---------------------------------------------------------------------===//
5013 // SSE3 - Arithmetic
5014 //===---------------------------------------------------------------------===//
5016 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5017 X86MemOperand x86memop, bit Is2Addr = 1> {
5018 def rr : I<0xD0, MRMSrcReg,
5019 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5021 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5022 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5023 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5024 def rm : I<0xD0, MRMSrcMem,
5025 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5027 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5028 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5029 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5032 let Predicates = [HasAVX] in {
5033 let ExeDomain = SSEPackedSingle in {
5034 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5035 f128mem, 0>, TB, XD, VEX_4V;
5036 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5037 f256mem, 0>, TB, XD, VEX_4V;
5039 let ExeDomain = SSEPackedDouble in {
5040 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5041 f128mem, 0>, TB, OpSize, VEX_4V;
5042 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5043 f256mem, 0>, TB, OpSize, VEX_4V;
5046 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5047 let ExeDomain = SSEPackedSingle in
5048 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5050 let ExeDomain = SSEPackedDouble in
5051 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5052 f128mem>, TB, OpSize;
5055 //===---------------------------------------------------------------------===//
5056 // SSE3 Instructions
5057 //===---------------------------------------------------------------------===//
5060 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5061 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5062 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5064 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5065 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5066 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5068 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5070 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5071 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5072 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5074 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5075 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5076 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5078 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5079 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5080 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5082 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5084 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5085 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5086 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5089 let Predicates = [HasAVX] in {
5090 let ExeDomain = SSEPackedSingle in {
5091 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5092 X86fhadd, 0>, VEX_4V;
5093 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5094 X86fhsub, 0>, VEX_4V;
5095 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5096 X86fhadd, 0>, VEX_4V;
5097 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5098 X86fhsub, 0>, VEX_4V;
5100 let ExeDomain = SSEPackedDouble in {
5101 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5102 X86fhadd, 0>, VEX_4V;
5103 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5104 X86fhsub, 0>, VEX_4V;
5105 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5106 X86fhadd, 0>, VEX_4V;
5107 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5108 X86fhsub, 0>, VEX_4V;
5112 let Constraints = "$src1 = $dst" in {
5113 let ExeDomain = SSEPackedSingle in {
5114 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5115 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5117 let ExeDomain = SSEPackedDouble in {
5118 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5119 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5123 //===---------------------------------------------------------------------===//
5124 // SSSE3 - Packed Absolute Instructions
5125 //===---------------------------------------------------------------------===//
5128 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5129 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5130 Intrinsic IntId128> {
5131 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5133 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5134 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5137 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5139 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5142 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5145 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5146 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5147 Intrinsic IntId256> {
5148 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5150 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5151 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5154 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5156 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5159 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5162 let Predicates = [HasAVX] in {
5163 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5164 int_x86_ssse3_pabs_b_128>, VEX;
5165 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5166 int_x86_ssse3_pabs_w_128>, VEX;
5167 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5168 int_x86_ssse3_pabs_d_128>, VEX;
5171 let Predicates = [HasAVX2] in {
5172 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5173 int_x86_avx2_pabs_b>, VEX;
5174 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5175 int_x86_avx2_pabs_w>, VEX;
5176 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5177 int_x86_avx2_pabs_d>, VEX;
5180 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5181 int_x86_ssse3_pabs_b_128>;
5182 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5183 int_x86_ssse3_pabs_w_128>;
5184 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5185 int_x86_ssse3_pabs_d_128>;
5187 //===---------------------------------------------------------------------===//
5188 // SSSE3 - Packed Binary Operator Instructions
5189 //===---------------------------------------------------------------------===//
5191 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5192 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5193 Intrinsic IntId128, bit Is2Addr = 1> {
5194 let isCommutable = 1 in
5195 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5196 (ins VR128:$src1, VR128:$src2),
5198 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5199 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5200 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5202 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5203 (ins VR128:$src1, i128mem:$src2),
5205 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5206 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5208 (IntId128 VR128:$src1,
5209 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5212 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5213 Intrinsic IntId256> {
5214 let isCommutable = 1 in
5215 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5216 (ins VR256:$src1, VR256:$src2),
5217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5218 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5220 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5221 (ins VR256:$src1, i256mem:$src2),
5222 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5224 (IntId256 VR256:$src1,
5225 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5228 let ImmT = NoImm, Predicates = [HasAVX] in {
5229 let isCommutable = 0 in {
5230 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5231 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5232 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5233 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5234 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5235 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5236 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5237 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5238 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5239 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5240 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5241 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5242 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5243 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5244 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5245 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5246 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5247 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5248 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5249 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5250 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5251 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5253 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5254 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5257 let ImmT = NoImm, Predicates = [HasAVX2] in {
5258 let isCommutable = 0 in {
5259 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5260 int_x86_avx2_phadd_w>, VEX_4V;
5261 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5262 int_x86_avx2_phadd_d>, VEX_4V;
5263 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5264 int_x86_avx2_phadd_sw>, VEX_4V;
5265 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5266 int_x86_avx2_phsub_w>, VEX_4V;
5267 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5268 int_x86_avx2_phsub_d>, VEX_4V;
5269 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5270 int_x86_avx2_phsub_sw>, VEX_4V;
5271 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5272 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5273 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5274 int_x86_avx2_pshuf_b>, VEX_4V;
5275 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5276 int_x86_avx2_psign_b>, VEX_4V;
5277 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5278 int_x86_avx2_psign_w>, VEX_4V;
5279 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5280 int_x86_avx2_psign_d>, VEX_4V;
5282 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5283 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5286 // None of these have i8 immediate fields.
5287 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5288 let isCommutable = 0 in {
5289 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5290 int_x86_ssse3_phadd_w_128>;
5291 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5292 int_x86_ssse3_phadd_d_128>;
5293 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5294 int_x86_ssse3_phadd_sw_128>;
5295 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5296 int_x86_ssse3_phsub_w_128>;
5297 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5298 int_x86_ssse3_phsub_d_128>;
5299 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5300 int_x86_ssse3_phsub_sw_128>;
5301 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5302 int_x86_ssse3_pmadd_ub_sw_128>;
5303 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5304 int_x86_ssse3_pshuf_b_128>;
5305 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5306 int_x86_ssse3_psign_b_128>;
5307 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5308 int_x86_ssse3_psign_w_128>;
5309 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5310 int_x86_ssse3_psign_d_128>;
5312 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5313 int_x86_ssse3_pmul_hr_sw_128>;
5316 let Predicates = [HasSSSE3] in {
5317 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5318 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5319 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5320 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5322 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5323 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5324 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5325 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5326 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5327 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5329 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5330 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5331 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5332 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5333 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5334 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5335 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5336 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5339 let Predicates = [HasAVX] in {
5340 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5341 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5342 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5343 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5345 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5346 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5347 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5348 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5349 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5350 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5352 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5353 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5354 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5355 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5356 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5357 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5358 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5359 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5362 let Predicates = [HasAVX2] in {
5363 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5364 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5365 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5366 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5367 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5368 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5370 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5371 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5372 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5373 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5374 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5375 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5376 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5377 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5380 //===---------------------------------------------------------------------===//
5381 // SSSE3 - Packed Align Instruction Patterns
5382 //===---------------------------------------------------------------------===//
5384 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5385 let neverHasSideEffects = 1 in {
5386 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5387 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5389 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5391 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5394 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5395 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5397 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5399 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5404 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5405 let neverHasSideEffects = 1 in {
5406 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5407 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5409 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5412 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5413 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5415 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5420 let Predicates = [HasAVX] in
5421 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5422 let Predicates = [HasAVX2] in
5423 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5424 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5425 defm PALIGN : ssse3_palign<"palignr">;
5427 let Predicates = [HasSSSE3] in {
5428 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5429 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5430 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5431 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5432 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5433 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5434 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5435 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5438 let Predicates = [HasAVX] in {
5439 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5440 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5441 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5442 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5443 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5444 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5445 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5446 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5449 //===---------------------------------------------------------------------===//
5450 // SSSE3 - Thread synchronization
5451 //===---------------------------------------------------------------------===//
5453 let usesCustomInserter = 1 in {
5454 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5455 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5456 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5457 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5460 let Uses = [EAX, ECX, EDX] in
5461 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5462 Requires<[HasSSE3]>;
5463 let Uses = [ECX, EAX] in
5464 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5465 Requires<[HasSSE3]>;
5467 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5468 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5470 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5471 Requires<[In32BitMode]>;
5472 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5473 Requires<[In64BitMode]>;
5475 //===----------------------------------------------------------------------===//
5476 // SSE4.1 - Packed Move with Sign/Zero Extend
5477 //===----------------------------------------------------------------------===//
5479 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5480 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5481 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5482 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5484 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5485 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5487 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5491 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5493 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5495 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5497 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5498 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5499 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5502 let Predicates = [HasAVX] in {
5503 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5505 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5507 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5509 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5511 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5513 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5517 let Predicates = [HasAVX2] in {
5518 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5519 int_x86_avx2_pmovsxbw>, VEX;
5520 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5521 int_x86_avx2_pmovsxwd>, VEX;
5522 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5523 int_x86_avx2_pmovsxdq>, VEX;
5524 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5525 int_x86_avx2_pmovzxbw>, VEX;
5526 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5527 int_x86_avx2_pmovzxwd>, VEX;
5528 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5529 int_x86_avx2_pmovzxdq>, VEX;
5532 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5533 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5534 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5535 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5536 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5537 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5539 let Predicates = [HasSSE41] in {
5540 // Common patterns involving scalar load.
5541 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5542 (PMOVSXBWrm addr:$src)>;
5543 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5544 (PMOVSXBWrm addr:$src)>;
5546 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5547 (PMOVSXWDrm addr:$src)>;
5548 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5549 (PMOVSXWDrm addr:$src)>;
5551 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5552 (PMOVSXDQrm addr:$src)>;
5553 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5554 (PMOVSXDQrm addr:$src)>;
5556 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5557 (PMOVZXBWrm addr:$src)>;
5558 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5559 (PMOVZXBWrm addr:$src)>;
5561 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5562 (PMOVZXWDrm addr:$src)>;
5563 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5564 (PMOVZXWDrm addr:$src)>;
5566 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5567 (PMOVZXDQrm addr:$src)>;
5568 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5569 (PMOVZXDQrm addr:$src)>;
5572 let Predicates = [HasAVX] in {
5573 // Common patterns involving scalar load.
5574 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5575 (VPMOVSXBWrm addr:$src)>;
5576 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5577 (VPMOVSXBWrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5580 (VPMOVSXWDrm addr:$src)>;
5581 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5582 (VPMOVSXWDrm addr:$src)>;
5584 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5585 (VPMOVSXDQrm addr:$src)>;
5586 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5587 (VPMOVSXDQrm addr:$src)>;
5589 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5590 (VPMOVZXBWrm addr:$src)>;
5591 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5592 (VPMOVZXBWrm addr:$src)>;
5594 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5595 (VPMOVZXWDrm addr:$src)>;
5596 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5597 (VPMOVZXWDrm addr:$src)>;
5599 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5600 (VPMOVZXDQrm addr:$src)>;
5601 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5602 (VPMOVZXDQrm addr:$src)>;
5606 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5607 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5608 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5609 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5611 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5612 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5614 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5618 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5620 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5621 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5622 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5624 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5625 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5627 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5631 let Predicates = [HasAVX] in {
5632 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5634 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5636 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5638 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5642 let Predicates = [HasAVX2] in {
5643 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5644 int_x86_avx2_pmovsxbd>, VEX;
5645 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5646 int_x86_avx2_pmovsxwq>, VEX;
5647 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5648 int_x86_avx2_pmovzxbd>, VEX;
5649 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5650 int_x86_avx2_pmovzxwq>, VEX;
5653 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5654 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5655 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5656 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5658 let Predicates = [HasSSE41] in {
5659 // Common patterns involving scalar load
5660 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5661 (PMOVSXBDrm addr:$src)>;
5662 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5663 (PMOVSXWQrm addr:$src)>;
5665 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5666 (PMOVZXBDrm addr:$src)>;
5667 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5668 (PMOVZXWQrm addr:$src)>;
5671 let Predicates = [HasAVX] in {
5672 // Common patterns involving scalar load
5673 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5674 (VPMOVSXBDrm addr:$src)>;
5675 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5676 (VPMOVSXWQrm addr:$src)>;
5678 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5679 (VPMOVZXBDrm addr:$src)>;
5680 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5681 (VPMOVZXWQrm addr:$src)>;
5684 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5685 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5686 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5687 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5689 // Expecting a i16 load any extended to i32 value.
5690 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5691 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5692 [(set VR128:$dst, (IntId (bitconvert
5693 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5697 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5699 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5700 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5701 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5703 // Expecting a i16 load any extended to i32 value.
5704 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5705 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5706 [(set VR256:$dst, (IntId (bitconvert
5707 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5711 let Predicates = [HasAVX] in {
5712 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5714 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5717 let Predicates = [HasAVX2] in {
5718 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5719 int_x86_avx2_pmovsxbq>, VEX;
5720 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5721 int_x86_avx2_pmovzxbq>, VEX;
5723 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5724 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5726 let Predicates = [HasSSE41] in {
5727 // Common patterns involving scalar load
5728 def : Pat<(int_x86_sse41_pmovsxbq
5729 (bitconvert (v4i32 (X86vzmovl
5730 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5731 (PMOVSXBQrm addr:$src)>;
5733 def : Pat<(int_x86_sse41_pmovzxbq
5734 (bitconvert (v4i32 (X86vzmovl
5735 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5736 (PMOVZXBQrm addr:$src)>;
5739 let Predicates = [HasAVX] in {
5740 // Common patterns involving scalar load
5741 def : Pat<(int_x86_sse41_pmovsxbq
5742 (bitconvert (v4i32 (X86vzmovl
5743 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5744 (VPMOVSXBQrm addr:$src)>;
5746 def : Pat<(int_x86_sse41_pmovzxbq
5747 (bitconvert (v4i32 (X86vzmovl
5748 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5749 (VPMOVZXBQrm addr:$src)>;
5752 //===----------------------------------------------------------------------===//
5753 // SSE4.1 - Extract Instructions
5754 //===----------------------------------------------------------------------===//
5756 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5757 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5758 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5759 (ins VR128:$src1, i32i8imm:$src2),
5760 !strconcat(OpcodeStr,
5761 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5762 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5764 let neverHasSideEffects = 1, mayStore = 1 in
5765 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5766 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5767 !strconcat(OpcodeStr,
5768 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5771 // There's an AssertZext in the way of writing the store pattern
5772 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5775 let Predicates = [HasAVX] in {
5776 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5777 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5778 (ins VR128:$src1, i32i8imm:$src2),
5779 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5782 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5785 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5786 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5787 let neverHasSideEffects = 1, mayStore = 1 in
5788 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5789 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5790 !strconcat(OpcodeStr,
5791 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5794 // There's an AssertZext in the way of writing the store pattern
5795 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5798 let Predicates = [HasAVX] in
5799 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5801 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5804 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5805 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5806 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5807 (ins VR128:$src1, i32i8imm:$src2),
5808 !strconcat(OpcodeStr,
5809 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5811 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5812 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5813 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5814 !strconcat(OpcodeStr,
5815 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5816 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5817 addr:$dst)]>, OpSize;
5820 let Predicates = [HasAVX] in
5821 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5823 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5825 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5826 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5827 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5828 (ins VR128:$src1, i32i8imm:$src2),
5829 !strconcat(OpcodeStr,
5830 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5832 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5833 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5834 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5835 !strconcat(OpcodeStr,
5836 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5837 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5838 addr:$dst)]>, OpSize, REX_W;
5841 let Predicates = [HasAVX] in
5842 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5844 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5846 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5848 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5849 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5850 (ins VR128:$src1, i32i8imm:$src2),
5851 !strconcat(OpcodeStr,
5852 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5854 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5856 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5857 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5858 !strconcat(OpcodeStr,
5859 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5860 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5861 addr:$dst)]>, OpSize;
5864 let ExeDomain = SSEPackedSingle in {
5865 let Predicates = [HasAVX] in {
5866 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5867 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5868 (ins VR128:$src1, i32i8imm:$src2),
5869 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5872 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5875 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5876 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5879 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5880 Requires<[HasSSE41]>;
5881 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5884 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5887 //===----------------------------------------------------------------------===//
5888 // SSE4.1 - Insert Instructions
5889 //===----------------------------------------------------------------------===//
5891 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5892 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5893 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5895 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5897 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5899 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5900 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5901 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5903 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5905 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5907 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5908 imm:$src3))]>, OpSize;
5911 let Predicates = [HasAVX] in
5912 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5913 let Constraints = "$src1 = $dst" in
5914 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5916 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5917 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5918 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5920 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5922 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5924 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5926 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5927 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5929 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5931 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5933 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5934 imm:$src3)))]>, OpSize;
5937 let Predicates = [HasAVX] in
5938 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5939 let Constraints = "$src1 = $dst" in
5940 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5942 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5943 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5944 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5946 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5948 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5950 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5952 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5953 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5955 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5957 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5959 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5960 imm:$src3)))]>, OpSize;
5963 let Predicates = [HasAVX] in
5964 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5965 let Constraints = "$src1 = $dst" in
5966 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5968 // insertps has a few different modes, there's the first two here below which
5969 // are optimized inserts that won't zero arbitrary elements in the destination
5970 // vector. The next one matches the intrinsic and could zero arbitrary elements
5971 // in the target vector.
5972 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5973 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5974 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5976 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5978 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5980 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5982 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5983 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5985 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5987 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5989 (X86insrtps VR128:$src1,
5990 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
5991 imm:$src3))]>, OpSize;
5994 let ExeDomain = SSEPackedSingle in {
5995 let Constraints = "$src1 = $dst" in
5996 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
5997 let Predicates = [HasAVX] in
5998 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6001 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6002 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6004 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6005 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6006 Requires<[HasSSE41]>;
6008 //===----------------------------------------------------------------------===//
6009 // SSE4.1 - Round Instructions
6010 //===----------------------------------------------------------------------===//
6012 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6013 X86MemOperand x86memop, RegisterClass RC,
6014 PatFrag mem_frag32, PatFrag mem_frag64,
6015 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6016 let ExeDomain = SSEPackedSingle in {
6017 // Intrinsic operation, reg.
6018 // Vector intrinsic operation, reg
6019 def PSr : SS4AIi8<opcps, MRMSrcReg,
6020 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6021 !strconcat(OpcodeStr,
6022 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6023 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6026 // Vector intrinsic operation, mem
6027 def PSm : SS4AIi8<opcps, MRMSrcMem,
6028 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6029 !strconcat(OpcodeStr,
6030 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6032 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6034 } // ExeDomain = SSEPackedSingle
6036 let ExeDomain = SSEPackedDouble in {
6037 // Vector intrinsic operation, reg
6038 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6039 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6040 !strconcat(OpcodeStr,
6041 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6042 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6045 // Vector intrinsic operation, mem
6046 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6047 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6048 !strconcat(OpcodeStr,
6049 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6051 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6053 } // ExeDomain = SSEPackedDouble
6056 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6059 Intrinsic F64Int, bit Is2Addr = 1> {
6060 let ExeDomain = GenericDomain in {
6062 def SSr : SS4AIi8<opcss, MRMSrcReg,
6063 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6065 !strconcat(OpcodeStr,
6066 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6067 !strconcat(OpcodeStr,
6068 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6071 // Intrinsic operation, reg.
6072 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6073 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6075 !strconcat(OpcodeStr,
6076 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6077 !strconcat(OpcodeStr,
6078 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6079 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6082 // Intrinsic operation, mem.
6083 def SSm : SS4AIi8<opcss, MRMSrcMem,
6084 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6086 !strconcat(OpcodeStr,
6087 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6088 !strconcat(OpcodeStr,
6089 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6091 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6095 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6096 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6098 !strconcat(OpcodeStr,
6099 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6100 !strconcat(OpcodeStr,
6101 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6104 // Intrinsic operation, reg.
6105 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6106 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2, i32i8imm:$src3),
6108 !strconcat(OpcodeStr,
6109 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6110 !strconcat(OpcodeStr,
6111 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6112 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6115 // Intrinsic operation, mem.
6116 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6117 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6119 !strconcat(OpcodeStr,
6120 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6121 !strconcat(OpcodeStr,
6122 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6124 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6126 } // ExeDomain = GenericDomain
6129 // FP round - roundss, roundps, roundsd, roundpd
6130 let Predicates = [HasAVX] in {
6132 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6133 memopv4f32, memopv2f64,
6134 int_x86_sse41_round_ps,
6135 int_x86_sse41_round_pd>, VEX;
6136 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6137 memopv8f32, memopv4f64,
6138 int_x86_avx_round_ps_256,
6139 int_x86_avx_round_pd_256>, VEX;
6140 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6141 int_x86_sse41_round_ss,
6142 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6144 def : Pat<(ffloor FR32:$src),
6145 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6146 def : Pat<(f64 (ffloor FR64:$src)),
6147 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6148 def : Pat<(f32 (fnearbyint FR32:$src)),
6149 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6150 def : Pat<(f64 (fnearbyint FR64:$src)),
6151 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6152 def : Pat<(f32 (fceil FR32:$src)),
6153 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6154 def : Pat<(f64 (fceil FR64:$src)),
6155 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6156 def : Pat<(f32 (frint FR32:$src)),
6157 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6158 def : Pat<(f64 (frint FR64:$src)),
6159 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6160 def : Pat<(f32 (ftrunc FR32:$src)),
6161 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6162 def : Pat<(f64 (ftrunc FR64:$src)),
6163 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6166 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6167 memopv4f32, memopv2f64,
6168 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6169 let Constraints = "$src1 = $dst" in
6170 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6171 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6173 def : Pat<(ffloor FR32:$src),
6174 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6175 def : Pat<(f64 (ffloor FR64:$src)),
6176 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6177 def : Pat<(f32 (fnearbyint FR32:$src)),
6178 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6179 def : Pat<(f64 (fnearbyint FR64:$src)),
6180 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6181 def : Pat<(f32 (fceil FR32:$src)),
6182 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6183 def : Pat<(f64 (fceil FR64:$src)),
6184 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6185 def : Pat<(f32 (frint FR32:$src)),
6186 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6187 def : Pat<(f64 (frint FR64:$src)),
6188 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6189 def : Pat<(f32 (ftrunc FR32:$src)),
6190 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6191 def : Pat<(f64 (ftrunc FR64:$src)),
6192 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6194 //===----------------------------------------------------------------------===//
6195 // SSE4.1 - Packed Bit Test
6196 //===----------------------------------------------------------------------===//
6198 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6199 // the intel intrinsic that corresponds to this.
6200 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6201 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6202 "vptest\t{$src2, $src1|$src1, $src2}",
6203 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6205 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6206 "vptest\t{$src2, $src1|$src1, $src2}",
6207 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6210 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6211 "vptest\t{$src2, $src1|$src1, $src2}",
6212 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6214 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6215 "vptest\t{$src2, $src1|$src1, $src2}",
6216 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6220 let Defs = [EFLAGS] in {
6221 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6222 "ptest\t{$src2, $src1|$src1, $src2}",
6223 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6225 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6226 "ptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6231 // The bit test instructions below are AVX only
6232 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6233 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6234 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6235 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6236 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6237 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6238 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6239 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6243 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6244 let ExeDomain = SSEPackedSingle in {
6245 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6246 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6248 let ExeDomain = SSEPackedDouble in {
6249 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6250 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6254 //===----------------------------------------------------------------------===//
6255 // SSE4.1 - Misc Instructions
6256 //===----------------------------------------------------------------------===//
6258 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6259 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6260 "popcnt{w}\t{$src, $dst|$dst, $src}",
6261 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6263 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6264 "popcnt{w}\t{$src, $dst|$dst, $src}",
6265 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6266 (implicit EFLAGS)]>, OpSize, XS;
6268 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6269 "popcnt{l}\t{$src, $dst|$dst, $src}",
6270 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6272 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6273 "popcnt{l}\t{$src, $dst|$dst, $src}",
6274 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6275 (implicit EFLAGS)]>, XS;
6277 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6278 "popcnt{q}\t{$src, $dst|$dst, $src}",
6279 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6281 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6282 "popcnt{q}\t{$src, $dst|$dst, $src}",
6283 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6284 (implicit EFLAGS)]>, XS;
6289 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6290 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6291 Intrinsic IntId128> {
6292 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6294 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6295 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6296 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6298 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6301 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6304 let Predicates = [HasAVX] in
6305 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6306 int_x86_sse41_phminposuw>, VEX;
6307 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6308 int_x86_sse41_phminposuw>;
6310 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6311 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6312 Intrinsic IntId128, bit Is2Addr = 1> {
6313 let isCommutable = 1 in
6314 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6315 (ins VR128:$src1, VR128:$src2),
6317 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6318 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6319 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6320 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6321 (ins VR128:$src1, i128mem:$src2),
6323 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6324 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6326 (IntId128 VR128:$src1,
6327 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6330 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6331 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6332 Intrinsic IntId256> {
6333 let isCommutable = 1 in
6334 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6335 (ins VR256:$src1, VR256:$src2),
6336 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6337 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6338 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6339 (ins VR256:$src1, i256mem:$src2),
6340 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6342 (IntId256 VR256:$src1,
6343 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6346 let Predicates = [HasAVX] in {
6347 let isCommutable = 0 in
6348 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6350 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6352 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6354 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6356 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6358 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6360 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6362 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6364 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6366 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6368 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6371 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6372 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6373 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6374 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6377 let Predicates = [HasAVX2] in {
6378 let isCommutable = 0 in
6379 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6380 int_x86_avx2_packusdw>, VEX_4V;
6381 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6382 int_x86_avx2_pcmpeq_q>, VEX_4V;
6383 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6384 int_x86_avx2_pmins_b>, VEX_4V;
6385 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6386 int_x86_avx2_pmins_d>, VEX_4V;
6387 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6388 int_x86_avx2_pminu_d>, VEX_4V;
6389 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6390 int_x86_avx2_pminu_w>, VEX_4V;
6391 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6392 int_x86_avx2_pmaxs_b>, VEX_4V;
6393 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6394 int_x86_avx2_pmaxs_d>, VEX_4V;
6395 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6396 int_x86_avx2_pmaxu_d>, VEX_4V;
6397 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6398 int_x86_avx2_pmaxu_w>, VEX_4V;
6399 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6400 int_x86_avx2_pmul_dq>, VEX_4V;
6402 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6403 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6404 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6405 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6408 let Constraints = "$src1 = $dst" in {
6409 let isCommutable = 0 in
6410 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6411 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6412 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6413 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6414 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6415 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6416 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6417 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6418 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6419 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6420 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6423 let Predicates = [HasSSE41] in {
6424 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6425 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6426 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6427 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6430 /// SS48I_binop_rm - Simple SSE41 binary operator.
6431 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6432 ValueType OpVT, bit Is2Addr = 1> {
6433 let isCommutable = 1 in
6434 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6435 (ins VR128:$src1, VR128:$src2),
6437 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6438 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6439 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6441 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6442 (ins VR128:$src1, i128mem:$src2),
6444 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6445 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6446 [(set VR128:$dst, (OpNode VR128:$src1,
6447 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6451 /// SS48I_binop_rm - Simple SSE41 binary operator.
6452 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6454 let isCommutable = 1 in
6455 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6456 (ins VR256:$src1, VR256:$src2),
6457 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6458 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6460 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6461 (ins VR256:$src1, i256mem:$src2),
6462 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6463 [(set VR256:$dst, (OpNode VR256:$src1,
6464 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6468 let Predicates = [HasAVX] in
6469 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6470 let Predicates = [HasAVX2] in
6471 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6472 let Constraints = "$src1 = $dst" in
6473 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6475 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6476 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6477 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6478 X86MemOperand x86memop, bit Is2Addr = 1> {
6479 let isCommutable = 1 in
6480 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6481 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6483 !strconcat(OpcodeStr,
6484 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6485 !strconcat(OpcodeStr,
6486 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6487 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6489 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6490 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6492 !strconcat(OpcodeStr,
6493 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6494 !strconcat(OpcodeStr,
6495 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6498 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6502 let Predicates = [HasAVX] in {
6503 let isCommutable = 0 in {
6504 let ExeDomain = SSEPackedSingle in {
6505 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6506 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6507 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6508 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6510 let ExeDomain = SSEPackedDouble in {
6511 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6512 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6513 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6514 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6516 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6517 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6518 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6519 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6521 let ExeDomain = SSEPackedSingle in
6522 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6523 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6524 let ExeDomain = SSEPackedDouble in
6525 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6526 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6527 let ExeDomain = SSEPackedSingle in
6528 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6529 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6532 let Predicates = [HasAVX2] in {
6533 let isCommutable = 0 in {
6534 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6535 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6536 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6537 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6541 let Constraints = "$src1 = $dst" in {
6542 let isCommutable = 0 in {
6543 let ExeDomain = SSEPackedSingle in
6544 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6545 VR128, memopv4f32, i128mem>;
6546 let ExeDomain = SSEPackedDouble in
6547 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6548 VR128, memopv2f64, i128mem>;
6549 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6550 VR128, memopv2i64, i128mem>;
6551 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6552 VR128, memopv2i64, i128mem>;
6554 let ExeDomain = SSEPackedSingle in
6555 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6556 VR128, memopv4f32, i128mem>;
6557 let ExeDomain = SSEPackedDouble in
6558 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6559 VR128, memopv2f64, i128mem>;
6562 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6563 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6564 RegisterClass RC, X86MemOperand x86memop,
6565 PatFrag mem_frag, Intrinsic IntId> {
6566 def rr : I<opc, MRMSrcReg, (outs RC:$dst),
6567 (ins RC:$src1, RC:$src2, RC:$src3),
6568 !strconcat(OpcodeStr,
6569 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6570 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6571 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6573 def rm : I<opc, MRMSrcMem, (outs RC:$dst),
6574 (ins RC:$src1, x86memop:$src2, RC:$src3),
6575 !strconcat(OpcodeStr,
6576 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6578 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6580 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6583 let Predicates = [HasAVX] in {
6584 let ExeDomain = SSEPackedDouble in {
6585 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6586 memopv2f64, int_x86_sse41_blendvpd>;
6587 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6588 memopv4f64, int_x86_avx_blendv_pd_256>;
6589 } // ExeDomain = SSEPackedDouble
6590 let ExeDomain = SSEPackedSingle in {
6591 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6592 memopv4f32, int_x86_sse41_blendvps>;
6593 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6594 memopv8f32, int_x86_avx_blendv_ps_256>;
6595 } // ExeDomain = SSEPackedSingle
6596 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6597 memopv2i64, int_x86_sse41_pblendvb>;
6600 let Predicates = [HasAVX2] in {
6601 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6602 memopv4i64, int_x86_avx2_pblendvb>;
6605 let Predicates = [HasAVX] in {
6606 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6607 (v16i8 VR128:$src2))),
6608 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6609 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6610 (v4i32 VR128:$src2))),
6611 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6612 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6613 (v4f32 VR128:$src2))),
6614 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6615 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6616 (v2i64 VR128:$src2))),
6617 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6618 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6619 (v2f64 VR128:$src2))),
6620 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6621 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6622 (v8i32 VR256:$src2))),
6623 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6624 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6625 (v8f32 VR256:$src2))),
6626 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6627 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6628 (v4i64 VR256:$src2))),
6629 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6630 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6631 (v4f64 VR256:$src2))),
6632 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6635 let Predicates = [HasAVX2] in {
6636 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6637 (v32i8 VR256:$src2))),
6638 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6641 /// SS41I_ternary_int - SSE 4.1 ternary operator
6642 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6643 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6645 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6646 (ins VR128:$src1, VR128:$src2),
6647 !strconcat(OpcodeStr,
6648 "\t{$src2, $dst|$dst, $src2}"),
6649 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6652 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6653 (ins VR128:$src1, i128mem:$src2),
6654 !strconcat(OpcodeStr,
6655 "\t{$src2, $dst|$dst, $src2}"),
6658 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6662 let ExeDomain = SSEPackedDouble in
6663 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6664 int_x86_sse41_blendvpd>;
6665 let ExeDomain = SSEPackedSingle in
6666 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6667 int_x86_sse41_blendvps>;
6668 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6669 int_x86_sse41_pblendvb>;
6671 let Predicates = [HasSSE41] in {
6672 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6673 (v16i8 VR128:$src2))),
6674 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6675 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6676 (v4i32 VR128:$src2))),
6677 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6678 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6679 (v4f32 VR128:$src2))),
6680 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6681 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6682 (v2i64 VR128:$src2))),
6683 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6684 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6685 (v2f64 VR128:$src2))),
6686 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6689 let Predicates = [HasAVX] in
6690 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6691 "vmovntdqa\t{$src, $dst|$dst, $src}",
6692 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6694 let Predicates = [HasAVX2] in
6695 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6696 "vmovntdqa\t{$src, $dst|$dst, $src}",
6697 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6699 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6700 "movntdqa\t{$src, $dst|$dst, $src}",
6701 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6704 //===----------------------------------------------------------------------===//
6705 // SSE4.2 - Compare Instructions
6706 //===----------------------------------------------------------------------===//
6708 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6709 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6710 Intrinsic IntId128, bit Is2Addr = 1> {
6711 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6712 (ins VR128:$src1, VR128:$src2),
6714 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6715 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6716 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6718 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6719 (ins VR128:$src1, i128mem:$src2),
6721 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6722 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6724 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6727 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6728 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6729 Intrinsic IntId256> {
6730 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6731 (ins VR256:$src1, VR256:$src2),
6732 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6733 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6735 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6736 (ins VR256:$src1, i256mem:$src2),
6737 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6739 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6742 let Predicates = [HasAVX] in {
6743 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6746 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6747 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6748 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6749 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6752 let Predicates = [HasAVX2] in {
6753 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6756 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6757 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6758 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6759 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6762 let Constraints = "$src1 = $dst" in
6763 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6765 let Predicates = [HasSSE42] in {
6766 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6767 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6768 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6769 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6772 //===----------------------------------------------------------------------===//
6773 // SSE4.2 - String/text Processing Instructions
6774 //===----------------------------------------------------------------------===//
6776 // Packed Compare Implicit Length Strings, Return Mask
6777 multiclass pseudo_pcmpistrm<string asm> {
6778 def REG : PseudoI<(outs VR128:$dst),
6779 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6780 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6782 def MEM : PseudoI<(outs VR128:$dst),
6783 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6784 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6785 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6788 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6789 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6790 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6793 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6794 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6795 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6796 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6798 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6799 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6800 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6803 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6804 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6805 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6806 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6808 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6809 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6810 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6813 // Packed Compare Explicit Length Strings, Return Mask
6814 multiclass pseudo_pcmpestrm<string asm> {
6815 def REG : PseudoI<(outs VR128:$dst),
6816 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6817 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6818 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6819 def MEM : PseudoI<(outs VR128:$dst),
6820 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6821 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6822 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6825 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6826 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6827 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6830 let Predicates = [HasAVX],
6831 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6832 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6833 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6834 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6836 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6837 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6838 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6841 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6842 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6843 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6844 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6846 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6847 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6848 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6851 // Packed Compare Implicit Length Strings, Return Index
6852 let Defs = [ECX, EFLAGS] in {
6853 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6854 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6855 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6856 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6857 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6858 (implicit EFLAGS)]>, OpSize;
6859 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6860 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6861 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6862 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6863 (implicit EFLAGS)]>, OpSize;
6867 let Predicates = [HasAVX] in {
6868 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6870 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6872 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6874 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6876 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6878 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6882 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6883 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6884 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6885 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6886 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6887 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6889 // Packed Compare Explicit Length Strings, Return Index
6890 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6891 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6892 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6893 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6894 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6895 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6896 (implicit EFLAGS)]>, OpSize;
6897 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6898 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6899 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6901 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6902 (implicit EFLAGS)]>, OpSize;
6906 let Predicates = [HasAVX] in {
6907 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6909 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6911 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6913 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6915 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6917 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6921 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6922 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6923 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6924 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6925 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6926 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6928 //===----------------------------------------------------------------------===//
6929 // SSE4.2 - CRC Instructions
6930 //===----------------------------------------------------------------------===//
6932 // No CRC instructions have AVX equivalents
6934 // crc intrinsic instruction
6935 // This set of instructions are only rm, the only difference is the size
6937 let Constraints = "$src1 = $dst" in {
6938 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6939 (ins GR32:$src1, i8mem:$src2),
6940 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6942 (int_x86_sse42_crc32_32_8 GR32:$src1,
6943 (load addr:$src2)))]>;
6944 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6945 (ins GR32:$src1, GR8:$src2),
6946 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6948 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6949 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6950 (ins GR32:$src1, i16mem:$src2),
6951 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6953 (int_x86_sse42_crc32_32_16 GR32:$src1,
6954 (load addr:$src2)))]>,
6956 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6957 (ins GR32:$src1, GR16:$src2),
6958 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6960 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6962 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6963 (ins GR32:$src1, i32mem:$src2),
6964 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6966 (int_x86_sse42_crc32_32_32 GR32:$src1,
6967 (load addr:$src2)))]>;
6968 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6969 (ins GR32:$src1, GR32:$src2),
6970 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6972 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6973 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6974 (ins GR64:$src1, i8mem:$src2),
6975 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6977 (int_x86_sse42_crc32_64_8 GR64:$src1,
6978 (load addr:$src2)))]>,
6980 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6981 (ins GR64:$src1, GR8:$src2),
6982 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6984 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6986 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6987 (ins GR64:$src1, i64mem:$src2),
6988 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6990 (int_x86_sse42_crc32_64_64 GR64:$src1,
6991 (load addr:$src2)))]>,
6993 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
6994 (ins GR64:$src1, GR64:$src2),
6995 "crc32{q} \t{$src2, $src1|$src1, $src2}",
6997 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7001 //===----------------------------------------------------------------------===//
7002 // AES-NI Instructions
7003 //===----------------------------------------------------------------------===//
7005 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7006 Intrinsic IntId128, bit Is2Addr = 1> {
7007 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7008 (ins VR128:$src1, VR128:$src2),
7010 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7011 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7012 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7014 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7015 (ins VR128:$src1, i128mem:$src2),
7017 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7018 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7020 (IntId128 VR128:$src1,
7021 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
7024 // Perform One Round of an AES Encryption/Decryption Flow
7025 let Predicates = [HasAVX, HasAES] in {
7026 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7027 int_x86_aesni_aesenc, 0>, VEX_4V;
7028 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7029 int_x86_aesni_aesenclast, 0>, VEX_4V;
7030 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7031 int_x86_aesni_aesdec, 0>, VEX_4V;
7032 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7033 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7036 let Constraints = "$src1 = $dst" in {
7037 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7038 int_x86_aesni_aesenc>;
7039 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7040 int_x86_aesni_aesenclast>;
7041 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7042 int_x86_aesni_aesdec>;
7043 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7044 int_x86_aesni_aesdeclast>;
7047 let Predicates = [HasAES] in {
7048 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
7049 (AESENCrr VR128:$src1, VR128:$src2)>;
7050 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
7051 (AESENCrm VR128:$src1, addr:$src2)>;
7052 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7053 (AESENCLASTrr VR128:$src1, VR128:$src2)>;
7054 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7055 (AESENCLASTrm VR128:$src1, addr:$src2)>;
7056 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7057 (AESDECrr VR128:$src1, VR128:$src2)>;
7058 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7059 (AESDECrm VR128:$src1, addr:$src2)>;
7060 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7061 (AESDECLASTrr VR128:$src1, VR128:$src2)>;
7062 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7063 (AESDECLASTrm VR128:$src1, addr:$src2)>;
7066 let Predicates = [HasAVX, HasAES], AddedComplexity = 20 in {
7067 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, VR128:$src2)),
7068 (VAESENCrr VR128:$src1, VR128:$src2)>;
7069 def : Pat<(v2i64 (int_x86_aesni_aesenc VR128:$src1, (memop addr:$src2))),
7070 (VAESENCrm VR128:$src1, addr:$src2)>;
7071 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, VR128:$src2)),
7072 (VAESENCLASTrr VR128:$src1, VR128:$src2)>;
7073 def : Pat<(v2i64 (int_x86_aesni_aesenclast VR128:$src1, (memop addr:$src2))),
7074 (VAESENCLASTrm VR128:$src1, addr:$src2)>;
7075 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, VR128:$src2)),
7076 (VAESDECrr VR128:$src1, VR128:$src2)>;
7077 def : Pat<(v2i64 (int_x86_aesni_aesdec VR128:$src1, (memop addr:$src2))),
7078 (VAESDECrm VR128:$src1, addr:$src2)>;
7079 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, VR128:$src2)),
7080 (VAESDECLASTrr VR128:$src1, VR128:$src2)>;
7081 def : Pat<(v2i64 (int_x86_aesni_aesdeclast VR128:$src1, (memop addr:$src2))),
7082 (VAESDECLASTrm VR128:$src1, addr:$src2)>;
7085 // Perform the AES InvMixColumn Transformation
7086 let Predicates = [HasAVX, HasAES] in {
7087 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7089 "vaesimc\t{$src1, $dst|$dst, $src1}",
7091 (int_x86_aesni_aesimc VR128:$src1))]>,
7093 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7094 (ins i128mem:$src1),
7095 "vaesimc\t{$src1, $dst|$dst, $src1}",
7097 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7100 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7102 "aesimc\t{$src1, $dst|$dst, $src1}",
7104 (int_x86_aesni_aesimc VR128:$src1))]>,
7106 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7107 (ins i128mem:$src1),
7108 "aesimc\t{$src1, $dst|$dst, $src1}",
7110 (int_x86_aesni_aesimc (bitconvert (memopv2i64 addr:$src1))))]>,
7113 // AES Round Key Generation Assist
7114 let Predicates = [HasAVX, HasAES] in {
7115 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7116 (ins VR128:$src1, i8imm:$src2),
7117 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7119 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7121 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7122 (ins i128mem:$src1, i8imm:$src2),
7123 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7125 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7129 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7130 (ins VR128:$src1, i8imm:$src2),
7131 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7133 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7135 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7136 (ins i128mem:$src1, i8imm:$src2),
7137 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7139 (int_x86_aesni_aeskeygenassist (bitconvert (memopv2i64 addr:$src1)),
7143 //===----------------------------------------------------------------------===//
7144 // CLMUL Instructions
7145 //===----------------------------------------------------------------------===//
7147 // Carry-less Multiplication instructions
7148 let neverHasSideEffects = 1 in {
7149 let Constraints = "$src1 = $dst" in {
7150 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7151 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7152 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7156 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7157 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7158 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7162 // AVX carry-less Multiplication instructions
7163 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7164 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7165 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7169 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7170 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7171 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7176 multiclass pclmul_alias<string asm, int immop> {
7177 def : InstAlias<!strconcat("pclmul", asm,
7178 "dq {$src, $dst|$dst, $src}"),
7179 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7181 def : InstAlias<!strconcat("pclmul", asm,
7182 "dq {$src, $dst|$dst, $src}"),
7183 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7185 def : InstAlias<!strconcat("vpclmul", asm,
7186 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7187 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7189 def : InstAlias<!strconcat("vpclmul", asm,
7190 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7191 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7193 defm : pclmul_alias<"hqhq", 0x11>;
7194 defm : pclmul_alias<"hqlq", 0x01>;
7195 defm : pclmul_alias<"lqhq", 0x10>;
7196 defm : pclmul_alias<"lqlq", 0x00>;
7198 //===----------------------------------------------------------------------===//
7200 //===----------------------------------------------------------------------===//
7202 //===----------------------------------------------------------------------===//
7203 // VBROADCAST - Load from memory and broadcast to all elements of the
7204 // destination operand
7206 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7207 X86MemOperand x86memop, Intrinsic Int> :
7208 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7209 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7210 [(set RC:$dst, (Int addr:$src))]>, VEX;
7212 // AVX2 adds register forms
7213 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7215 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7216 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7217 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7219 let ExeDomain = SSEPackedSingle in {
7220 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7221 int_x86_avx_vbroadcast_ss>;
7222 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7223 int_x86_avx_vbroadcast_ss_256>;
7225 let ExeDomain = SSEPackedDouble in
7226 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7227 int_x86_avx_vbroadcast_sd_256>;
7228 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7229 int_x86_avx_vbroadcastf128_pd_256>;
7231 let ExeDomain = SSEPackedSingle in {
7232 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7233 int_x86_avx2_vbroadcast_ss_ps>;
7234 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7235 int_x86_avx2_vbroadcast_ss_ps_256>;
7237 let ExeDomain = SSEPackedDouble in
7238 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7239 int_x86_avx2_vbroadcast_sd_pd_256>;
7241 let Predicates = [HasAVX2] in
7242 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7243 int_x86_avx2_vbroadcasti128>;
7245 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7246 (VBROADCASTF128 addr:$src)>;
7249 //===----------------------------------------------------------------------===//
7250 // VINSERTF128 - Insert packed floating-point values
7252 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7253 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7254 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7255 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7258 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7259 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7260 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7264 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7265 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7266 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7267 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7268 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7269 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7271 //===----------------------------------------------------------------------===//
7272 // VEXTRACTF128 - Extract packed floating-point values
7274 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7275 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7276 (ins VR256:$src1, i8imm:$src2),
7277 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7280 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7281 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7282 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7286 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7287 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7288 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7289 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7290 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7291 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7293 //===----------------------------------------------------------------------===//
7294 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7296 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7297 Intrinsic IntLd, Intrinsic IntLd256,
7298 Intrinsic IntSt, Intrinsic IntSt256> {
7299 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7300 (ins VR128:$src1, f128mem:$src2),
7301 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7302 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7304 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7305 (ins VR256:$src1, f256mem:$src2),
7306 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7307 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7309 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7310 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7311 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7312 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7313 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7314 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7315 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7316 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7319 let ExeDomain = SSEPackedSingle in
7320 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7321 int_x86_avx_maskload_ps,
7322 int_x86_avx_maskload_ps_256,
7323 int_x86_avx_maskstore_ps,
7324 int_x86_avx_maskstore_ps_256>;
7325 let ExeDomain = SSEPackedDouble in
7326 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7327 int_x86_avx_maskload_pd,
7328 int_x86_avx_maskload_pd_256,
7329 int_x86_avx_maskstore_pd,
7330 int_x86_avx_maskstore_pd_256>;
7332 //===----------------------------------------------------------------------===//
7333 // VPERMIL - Permute Single and Double Floating-Point Values
7335 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7336 RegisterClass RC, X86MemOperand x86memop_f,
7337 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7338 Intrinsic IntVar, Intrinsic IntImm> {
7339 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7340 (ins RC:$src1, RC:$src2),
7341 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7342 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7343 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7344 (ins RC:$src1, x86memop_i:$src2),
7345 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7346 [(set RC:$dst, (IntVar RC:$src1,
7347 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7349 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7350 (ins RC:$src1, i8imm:$src2),
7351 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7352 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7353 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7354 (ins x86memop_f:$src1, i8imm:$src2),
7355 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7356 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7359 let ExeDomain = SSEPackedSingle in {
7360 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7361 memopv4f32, memopv2i64,
7362 int_x86_avx_vpermilvar_ps,
7363 int_x86_avx_vpermil_ps>;
7364 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7365 memopv8f32, memopv4i64,
7366 int_x86_avx_vpermilvar_ps_256,
7367 int_x86_avx_vpermil_ps_256>;
7369 let ExeDomain = SSEPackedDouble in {
7370 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7371 memopv2f64, memopv2i64,
7372 int_x86_avx_vpermilvar_pd,
7373 int_x86_avx_vpermil_pd>;
7374 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7375 memopv4f64, memopv4i64,
7376 int_x86_avx_vpermilvar_pd_256,
7377 int_x86_avx_vpermil_pd_256>;
7380 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7381 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7382 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7383 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7384 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7385 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7386 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7387 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7388 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7389 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7390 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7391 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7392 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7394 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7395 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7396 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7398 //===----------------------------------------------------------------------===//
7399 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7401 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7402 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7403 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7404 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7407 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7408 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7409 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7413 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7414 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7415 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7416 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7417 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7418 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7420 def : Pat<(int_x86_avx_vperm2f128_ps_256
7421 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7422 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7423 def : Pat<(int_x86_avx_vperm2f128_pd_256
7424 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7425 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7426 def : Pat<(int_x86_avx_vperm2f128_si_256
7427 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7428 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7430 //===----------------------------------------------------------------------===//
7431 // VZERO - Zero YMM registers
7433 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7434 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7435 // Zero All YMM registers
7436 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7437 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7439 // Zero Upper bits of YMM registers
7440 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7441 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7444 //===----------------------------------------------------------------------===//
7445 // Half precision conversion instructions
7446 //===----------------------------------------------------------------------===//
7447 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7448 let Predicates = [HasAVX, HasF16C] in {
7449 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7450 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7451 [(set RC:$dst, (Int VR128:$src))]>,
7453 let neverHasSideEffects = 1, mayLoad = 1 in
7454 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7455 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7459 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7460 let Predicates = [HasAVX, HasF16C] in {
7461 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7462 (ins RC:$src1, i32i8imm:$src2),
7463 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7464 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7466 let neverHasSideEffects = 1, mayLoad = 1 in
7467 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7468 (ins RC:$src1, i32i8imm:$src2),
7469 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7474 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7475 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7476 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7477 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7479 //===----------------------------------------------------------------------===//
7480 // AVX2 Instructions
7481 //===----------------------------------------------------------------------===//
7483 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7484 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7485 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7486 X86MemOperand x86memop> {
7487 let isCommutable = 1 in
7488 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7489 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7490 !strconcat(OpcodeStr,
7491 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7492 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7494 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7495 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7496 !strconcat(OpcodeStr,
7497 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7500 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7504 let isCommutable = 0 in {
7505 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7506 VR128, memopv2i64, i128mem>;
7507 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7508 VR256, memopv4i64, i256mem>;
7511 //===----------------------------------------------------------------------===//
7512 // VPBROADCAST - Load from memory and broadcast to all elements of the
7513 // destination operand
7515 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7516 X86MemOperand x86memop, PatFrag ld_frag,
7517 Intrinsic Int128, Intrinsic Int256> {
7518 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7519 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7520 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7521 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7522 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7524 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7525 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7526 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7527 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7528 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7529 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7531 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7534 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7535 int_x86_avx2_pbroadcastb_128,
7536 int_x86_avx2_pbroadcastb_256>;
7537 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7538 int_x86_avx2_pbroadcastw_128,
7539 int_x86_avx2_pbroadcastw_256>;
7540 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7541 int_x86_avx2_pbroadcastd_128,
7542 int_x86_avx2_pbroadcastd_256>;
7543 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7544 int_x86_avx2_pbroadcastq_128,
7545 int_x86_avx2_pbroadcastq_256>;
7547 let Predicates = [HasAVX2] in {
7548 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7549 (VPBROADCASTBrm addr:$src)>;
7550 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7551 (VPBROADCASTBYrm addr:$src)>;
7552 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7553 (VPBROADCASTWrm addr:$src)>;
7554 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7555 (VPBROADCASTWYrm addr:$src)>;
7556 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7557 (VPBROADCASTDrm addr:$src)>;
7558 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7559 (VPBROADCASTDYrm addr:$src)>;
7560 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7561 (VPBROADCASTQrm addr:$src)>;
7562 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7563 (VPBROADCASTQYrm addr:$src)>;
7566 // AVX1 broadcast patterns
7567 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7568 (VBROADCASTSSYrm addr:$src)>;
7569 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7570 (VBROADCASTSDrm addr:$src)>;
7571 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7572 (VBROADCASTSSYrm addr:$src)>;
7573 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7574 (VBROADCASTSDrm addr:$src)>;
7576 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7577 (VBROADCASTSSrm addr:$src)>;
7578 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7579 (VBROADCASTSSrm addr:$src)>;
7581 //===----------------------------------------------------------------------===//
7582 // VPERM - Permute instructions
7585 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7587 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7588 (ins VR256:$src1, VR256:$src2),
7589 !strconcat(OpcodeStr,
7590 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7591 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7592 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7593 (ins VR256:$src1, i256mem:$src2),
7594 !strconcat(OpcodeStr,
7595 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7596 [(set VR256:$dst, (Int VR256:$src1,
7597 (bitconvert (mem_frag addr:$src2))))]>,
7601 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7602 let ExeDomain = SSEPackedSingle in
7603 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7605 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7607 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7608 (ins VR256:$src1, i8imm:$src2),
7609 !strconcat(OpcodeStr,
7610 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7611 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7612 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7613 (ins i256mem:$src1, i8imm:$src2),
7614 !strconcat(OpcodeStr,
7615 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7616 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7620 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7622 let ExeDomain = SSEPackedDouble in
7623 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7626 //===----------------------------------------------------------------------===//
7627 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7629 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7630 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7631 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7633 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7635 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7636 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7637 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7639 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7643 let Predicates = [HasAVX2] in {
7644 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7645 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7646 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7647 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7648 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7649 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7650 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7651 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7653 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7655 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7656 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7657 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7658 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7659 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7661 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7662 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7664 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7668 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7669 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7670 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7671 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7672 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7673 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7674 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7675 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7676 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7677 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7678 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7679 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7681 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7682 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7683 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7684 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7685 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7686 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7687 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7688 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7689 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7690 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7691 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7692 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7693 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7694 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7695 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7696 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7697 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7698 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7701 //===----------------------------------------------------------------------===//
7702 // VINSERTI128 - Insert packed integer values
7704 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7705 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7706 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7708 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7710 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7711 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7712 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7714 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7715 imm:$src3))]>, VEX_4V;
7717 let Predicates = [HasAVX2] in {
7718 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7720 (VINSERTI128rr VR256:$src1, VR128:$src2,
7721 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7722 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7724 (VINSERTI128rr VR256:$src1, VR128:$src2,
7725 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7726 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7728 (VINSERTI128rr VR256:$src1, VR128:$src2,
7729 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7730 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7732 (VINSERTI128rr VR256:$src1, VR128:$src2,
7733 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7737 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7739 (VINSERTF128rr VR256:$src1, VR128:$src2,
7740 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7741 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7743 (VINSERTF128rr VR256:$src1, VR128:$src2,
7744 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7745 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7747 (VINSERTF128rr VR256:$src1, VR128:$src2,
7748 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7749 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7751 (VINSERTF128rr VR256:$src1, VR128:$src2,
7752 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7753 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7755 (VINSERTF128rr VR256:$src1, VR128:$src2,
7756 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7757 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7759 (VINSERTF128rr VR256:$src1, VR128:$src2,
7760 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7762 //===----------------------------------------------------------------------===//
7763 // VEXTRACTI128 - Extract packed integer values
7765 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7766 (ins VR256:$src1, i8imm:$src2),
7767 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7769 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7771 let neverHasSideEffects = 1, mayStore = 1 in
7772 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7773 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7774 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7776 let Predicates = [HasAVX2] in {
7777 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7778 (v2i64 (VEXTRACTI128rr
7779 (v4i64 VR256:$src1),
7780 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7781 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7782 (v4i32 (VEXTRACTI128rr
7783 (v8i32 VR256:$src1),
7784 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7785 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7786 (v8i16 (VEXTRACTI128rr
7787 (v16i16 VR256:$src1),
7788 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7789 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7790 (v16i8 (VEXTRACTI128rr
7791 (v32i8 VR256:$src1),
7792 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7796 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7797 (v4f32 (VEXTRACTF128rr
7798 (v8f32 VR256:$src1),
7799 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7800 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7801 (v2f64 (VEXTRACTF128rr
7802 (v4f64 VR256:$src1),
7803 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7804 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7805 (v2i64 (VEXTRACTF128rr
7806 (v4i64 VR256:$src1),
7807 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7808 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7809 (v4i32 (VEXTRACTF128rr
7810 (v8i32 VR256:$src1),
7811 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7812 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7813 (v8i16 (VEXTRACTF128rr
7814 (v16i16 VR256:$src1),
7815 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7816 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7817 (v16i8 (VEXTRACTF128rr
7818 (v32i8 VR256:$src1),
7819 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7821 //===----------------------------------------------------------------------===//
7822 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7824 multiclass avx2_pmovmask<string OpcodeStr,
7825 Intrinsic IntLd128, Intrinsic IntLd256,
7826 Intrinsic IntSt128, Intrinsic IntSt256> {
7827 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7828 (ins VR128:$src1, i128mem:$src2),
7829 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7830 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7831 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7832 (ins VR256:$src1, i256mem:$src2),
7833 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7834 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7835 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7836 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7837 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7838 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7839 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7840 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7841 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7842 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7845 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7846 int_x86_avx2_maskload_d,
7847 int_x86_avx2_maskload_d_256,
7848 int_x86_avx2_maskstore_d,
7849 int_x86_avx2_maskstore_d_256>;
7850 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7851 int_x86_avx2_maskload_q,
7852 int_x86_avx2_maskload_q_256,
7853 int_x86_avx2_maskstore_q,
7854 int_x86_avx2_maskstore_q_256>, VEX_W;
7857 //===----------------------------------------------------------------------===//
7858 // Variable Bit Shifts
7860 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7861 ValueType vt128, ValueType vt256> {
7862 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7863 (ins VR128:$src1, VR128:$src2),
7864 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7866 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7868 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7869 (ins VR128:$src1, i128mem:$src2),
7870 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7872 (vt128 (OpNode VR128:$src1,
7873 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7875 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7876 (ins VR256:$src1, VR256:$src2),
7877 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7879 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7881 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7882 (ins VR256:$src1, i256mem:$src2),
7883 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7885 (vt256 (OpNode VR256:$src1,
7886 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7890 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7891 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7892 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7893 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7894 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;