1 //====- X86InstrSSE.td - Describe the X86 Instruction Set --*- tablegen -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file describes the X86 SSE instruction set, defining the instructions,
11 // and properties of the instructions which are needed for code generation,
12 // machine code emission, and analysis.
14 //===----------------------------------------------------------------------===//
17 //===----------------------------------------------------------------------===//
18 // SSE 1 & 2 Instructions Classes
19 //===----------------------------------------------------------------------===//
21 /// sse12_fp_scalar - SSE 1 & 2 scalar instructions class
22 multiclass sse12_fp_scalar<bits<8> opc, string OpcodeStr, SDNode OpNode,
23 RegisterClass RC, X86MemOperand x86memop,
25 let isCommutable = 1 in {
26 def rr : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
28 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
29 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
30 [(set RC:$dst, (OpNode RC:$src1, RC:$src2))]>;
32 def rm : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
34 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
35 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
36 [(set RC:$dst, (OpNode RC:$src1, (load addr:$src2)))]>;
39 /// sse12_fp_scalar_int - SSE 1 & 2 scalar instructions intrinsics class
40 multiclass sse12_fp_scalar_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
41 string asm, string SSEVer, string FPSizeStr,
42 Operand memopr, ComplexPattern mem_cpat,
44 def rr_Int : SI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
46 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
47 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
48 [(set RC:$dst, (!cast<Intrinsic>(
49 !strconcat("int_x86_sse", SSEVer, "_", OpcodeStr, FPSizeStr))
50 RC:$src1, RC:$src2))]>;
51 def rm_Int : SI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, memopr:$src2),
53 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
54 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
55 [(set RC:$dst, (!cast<Intrinsic>(!strconcat("int_x86_sse",
56 SSEVer, "_", OpcodeStr, FPSizeStr))
57 RC:$src1, mem_cpat:$src2))]>;
60 /// sse12_fp_packed - SSE 1 & 2 packed instructions class
61 multiclass sse12_fp_packed<bits<8> opc, string OpcodeStr, SDNode OpNode,
62 RegisterClass RC, ValueType vt,
63 X86MemOperand x86memop, PatFrag mem_frag,
64 Domain d, bit Is2Addr = 1> {
65 let isCommutable = 1 in
66 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
68 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
69 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
70 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))], d>;
72 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
74 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
75 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
76 [(set RC:$dst, (OpNode RC:$src1, (mem_frag addr:$src2)))], d>;
79 /// sse12_fp_packed_logical_rm - SSE 1 & 2 packed instructions class
80 multiclass sse12_fp_packed_logical_rm<bits<8> opc, RegisterClass RC, Domain d,
81 string OpcodeStr, X86MemOperand x86memop,
82 list<dag> pat_rr, list<dag> pat_rm,
84 bit rr_hasSideEffects = 0> {
85 let isCommutable = 1, neverHasSideEffects = rr_hasSideEffects in
86 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
88 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
89 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
91 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
93 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
94 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
98 /// sse12_fp_packed_int - SSE 1 & 2 packed instructions intrinsics class
99 multiclass sse12_fp_packed_int<bits<8> opc, string OpcodeStr, RegisterClass RC,
100 string asm, string SSEVer, string FPSizeStr,
101 X86MemOperand x86memop, PatFrag mem_frag,
102 Domain d, bit Is2Addr = 1> {
103 def rr_Int : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
105 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
106 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
107 [(set RC:$dst, (!cast<Intrinsic>(
108 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
109 RC:$src1, RC:$src2))], d>;
110 def rm_Int : PI<opc, MRMSrcMem, (outs RC:$dst), (ins RC:$src1,x86memop:$src2),
112 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
113 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
114 [(set RC:$dst, (!cast<Intrinsic>(
115 !strconcat("int_x86_", SSEVer, "_", OpcodeStr, FPSizeStr))
116 RC:$src1, (mem_frag addr:$src2)))], d>;
119 //===----------------------------------------------------------------------===//
120 // Non-instruction patterns
121 //===----------------------------------------------------------------------===//
123 // A vector extract of the first f32/f64 position is a subregister copy
124 def : Pat<(f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
125 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
126 def : Pat<(f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
127 (f64 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
129 // A 128-bit subvector extract from the first 256-bit vector position
130 // is a subregister copy that needs no instruction.
131 def : Pat<(v4i32 (extract_subvector (v8i32 VR256:$src), (i32 0))),
132 (v4i32 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_xmm))>;
133 def : Pat<(v4f32 (extract_subvector (v8f32 VR256:$src), (i32 0))),
134 (v4f32 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_xmm))>;
136 def : Pat<(v2i64 (extract_subvector (v4i64 VR256:$src), (i32 0))),
137 (v2i64 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_xmm))>;
138 def : Pat<(v2f64 (extract_subvector (v4f64 VR256:$src), (i32 0))),
139 (v2f64 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_xmm))>;
141 def : Pat<(v8i16 (extract_subvector (v16i16 VR256:$src), (i32 0))),
142 (v8i16 (EXTRACT_SUBREG (v16i16 VR256:$src), sub_xmm))>;
143 def : Pat<(v16i8 (extract_subvector (v32i8 VR256:$src), (i32 0))),
144 (v16i8 (EXTRACT_SUBREG (v32i8 VR256:$src), sub_xmm))>;
146 // A 128-bit subvector insert to the first 256-bit vector position
147 // is a subregister copy that needs no instruction.
148 def : Pat<(insert_subvector undef, (v2i64 VR128:$src), (i32 0)),
149 (INSERT_SUBREG (v4i64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
150 def : Pat<(insert_subvector undef, (v2f64 VR128:$src), (i32 0)),
151 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
152 def : Pat<(insert_subvector undef, (v4i32 VR128:$src), (i32 0)),
153 (INSERT_SUBREG (v8i32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
154 def : Pat<(insert_subvector undef, (v4f32 VR128:$src), (i32 0)),
155 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
156 def : Pat<(insert_subvector undef, (v8i16 VR128:$src), (i32 0)),
157 (INSERT_SUBREG (v16i16 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
158 def : Pat<(insert_subvector undef, (v16i8 VR128:$src), (i32 0)),
159 (INSERT_SUBREG (v32i8 (IMPLICIT_DEF)), VR128:$src, sub_xmm)>;
161 // Implicitly promote a 32-bit scalar to a vector.
162 def : Pat<(v4f32 (scalar_to_vector FR32:$src)),
163 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
164 def : Pat<(v8f32 (scalar_to_vector FR32:$src)),
165 (INSERT_SUBREG (v8f32 (IMPLICIT_DEF)), FR32:$src, sub_ss)>;
166 // Implicitly promote a 64-bit scalar to a vector.
167 def : Pat<(v2f64 (scalar_to_vector FR64:$src)),
168 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
169 def : Pat<(v4f64 (scalar_to_vector FR64:$src)),
170 (INSERT_SUBREG (v4f64 (IMPLICIT_DEF)), FR64:$src, sub_sd)>;
172 // Bitcasts between 128-bit vector types. Return the original type since
173 // no instruction is needed for the conversion
174 let Predicates = [HasXMMInt] in {
175 def : Pat<(v2i64 (bitconvert (v4i32 VR128:$src))), (v2i64 VR128:$src)>;
176 def : Pat<(v2i64 (bitconvert (v8i16 VR128:$src))), (v2i64 VR128:$src)>;
177 def : Pat<(v2i64 (bitconvert (v16i8 VR128:$src))), (v2i64 VR128:$src)>;
178 def : Pat<(v2i64 (bitconvert (v2f64 VR128:$src))), (v2i64 VR128:$src)>;
179 def : Pat<(v2i64 (bitconvert (v4f32 VR128:$src))), (v2i64 VR128:$src)>;
180 def : Pat<(v4i32 (bitconvert (v2i64 VR128:$src))), (v4i32 VR128:$src)>;
181 def : Pat<(v4i32 (bitconvert (v8i16 VR128:$src))), (v4i32 VR128:$src)>;
182 def : Pat<(v4i32 (bitconvert (v16i8 VR128:$src))), (v4i32 VR128:$src)>;
183 def : Pat<(v4i32 (bitconvert (v2f64 VR128:$src))), (v4i32 VR128:$src)>;
184 def : Pat<(v4i32 (bitconvert (v4f32 VR128:$src))), (v4i32 VR128:$src)>;
185 def : Pat<(v8i16 (bitconvert (v2i64 VR128:$src))), (v8i16 VR128:$src)>;
186 def : Pat<(v8i16 (bitconvert (v4i32 VR128:$src))), (v8i16 VR128:$src)>;
187 def : Pat<(v8i16 (bitconvert (v16i8 VR128:$src))), (v8i16 VR128:$src)>;
188 def : Pat<(v8i16 (bitconvert (v2f64 VR128:$src))), (v8i16 VR128:$src)>;
189 def : Pat<(v8i16 (bitconvert (v4f32 VR128:$src))), (v8i16 VR128:$src)>;
190 def : Pat<(v16i8 (bitconvert (v2i64 VR128:$src))), (v16i8 VR128:$src)>;
191 def : Pat<(v16i8 (bitconvert (v4i32 VR128:$src))), (v16i8 VR128:$src)>;
192 def : Pat<(v16i8 (bitconvert (v8i16 VR128:$src))), (v16i8 VR128:$src)>;
193 def : Pat<(v16i8 (bitconvert (v2f64 VR128:$src))), (v16i8 VR128:$src)>;
194 def : Pat<(v16i8 (bitconvert (v4f32 VR128:$src))), (v16i8 VR128:$src)>;
195 def : Pat<(v4f32 (bitconvert (v2i64 VR128:$src))), (v4f32 VR128:$src)>;
196 def : Pat<(v4f32 (bitconvert (v4i32 VR128:$src))), (v4f32 VR128:$src)>;
197 def : Pat<(v4f32 (bitconvert (v8i16 VR128:$src))), (v4f32 VR128:$src)>;
198 def : Pat<(v4f32 (bitconvert (v16i8 VR128:$src))), (v4f32 VR128:$src)>;
199 def : Pat<(v4f32 (bitconvert (v2f64 VR128:$src))), (v4f32 VR128:$src)>;
200 def : Pat<(v2f64 (bitconvert (v2i64 VR128:$src))), (v2f64 VR128:$src)>;
201 def : Pat<(v2f64 (bitconvert (v4i32 VR128:$src))), (v2f64 VR128:$src)>;
202 def : Pat<(v2f64 (bitconvert (v8i16 VR128:$src))), (v2f64 VR128:$src)>;
203 def : Pat<(v2f64 (bitconvert (v16i8 VR128:$src))), (v2f64 VR128:$src)>;
204 def : Pat<(v2f64 (bitconvert (v4f32 VR128:$src))), (v2f64 VR128:$src)>;
207 // Bitcasts between 256-bit vector types. Return the original type since
208 // no instruction is needed for the conversion
209 let Predicates = [HasAVX] in {
210 def : Pat<(v4f64 (bitconvert (v8f32 VR256:$src))), (v4f64 VR256:$src)>;
211 def : Pat<(v4f64 (bitconvert (v8i32 VR256:$src))), (v4f64 VR256:$src)>;
212 def : Pat<(v4f64 (bitconvert (v4i64 VR256:$src))), (v4f64 VR256:$src)>;
213 def : Pat<(v4f64 (bitconvert (v16i16 VR256:$src))), (v4f64 VR256:$src)>;
214 def : Pat<(v4f64 (bitconvert (v32i8 VR256:$src))), (v4f64 VR256:$src)>;
215 def : Pat<(v8f32 (bitconvert (v8i32 VR256:$src))), (v8f32 VR256:$src)>;
216 def : Pat<(v8f32 (bitconvert (v4i64 VR256:$src))), (v8f32 VR256:$src)>;
217 def : Pat<(v8f32 (bitconvert (v4f64 VR256:$src))), (v8f32 VR256:$src)>;
218 def : Pat<(v8f32 (bitconvert (v32i8 VR256:$src))), (v8f32 VR256:$src)>;
219 def : Pat<(v8f32 (bitconvert (v16i16 VR256:$src))), (v8f32 VR256:$src)>;
220 def : Pat<(v4i64 (bitconvert (v8f32 VR256:$src))), (v4i64 VR256:$src)>;
221 def : Pat<(v4i64 (bitconvert (v8i32 VR256:$src))), (v4i64 VR256:$src)>;
222 def : Pat<(v4i64 (bitconvert (v4f64 VR256:$src))), (v4i64 VR256:$src)>;
223 def : Pat<(v4i64 (bitconvert (v32i8 VR256:$src))), (v4i64 VR256:$src)>;
224 def : Pat<(v4i64 (bitconvert (v16i16 VR256:$src))), (v4i64 VR256:$src)>;
225 def : Pat<(v32i8 (bitconvert (v4f64 VR256:$src))), (v32i8 VR256:$src)>;
226 def : Pat<(v32i8 (bitconvert (v4i64 VR256:$src))), (v32i8 VR256:$src)>;
227 def : Pat<(v32i8 (bitconvert (v8f32 VR256:$src))), (v32i8 VR256:$src)>;
228 def : Pat<(v32i8 (bitconvert (v8i32 VR256:$src))), (v32i8 VR256:$src)>;
229 def : Pat<(v32i8 (bitconvert (v16i16 VR256:$src))), (v32i8 VR256:$src)>;
230 def : Pat<(v8i32 (bitconvert (v32i8 VR256:$src))), (v8i32 VR256:$src)>;
231 def : Pat<(v8i32 (bitconvert (v16i16 VR256:$src))), (v8i32 VR256:$src)>;
232 def : Pat<(v8i32 (bitconvert (v8f32 VR256:$src))), (v8i32 VR256:$src)>;
233 def : Pat<(v8i32 (bitconvert (v4i64 VR256:$src))), (v8i32 VR256:$src)>;
234 def : Pat<(v8i32 (bitconvert (v4f64 VR256:$src))), (v8i32 VR256:$src)>;
235 def : Pat<(v16i16 (bitconvert (v8f32 VR256:$src))), (v16i16 VR256:$src)>;
236 def : Pat<(v16i16 (bitconvert (v8i32 VR256:$src))), (v16i16 VR256:$src)>;
237 def : Pat<(v16i16 (bitconvert (v4i64 VR256:$src))), (v16i16 VR256:$src)>;
238 def : Pat<(v16i16 (bitconvert (v4f64 VR256:$src))), (v16i16 VR256:$src)>;
239 def : Pat<(v16i16 (bitconvert (v32i8 VR256:$src))), (v16i16 VR256:$src)>;
242 // Alias instructions that map fld0 to pxor for sse.
243 // This is expanded by ExpandPostRAPseudos.
244 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
246 def FsFLD0SS : I<0, Pseudo, (outs FR32:$dst), (ins), "",
247 [(set FR32:$dst, fp32imm0)]>, Requires<[HasXMM]>;
248 def FsFLD0SD : I<0, Pseudo, (outs FR64:$dst), (ins), "",
249 [(set FR64:$dst, fpimm0)]>, Requires<[HasXMMInt]>;
252 //===----------------------------------------------------------------------===//
253 // AVX & SSE - Zero/One Vectors
254 //===----------------------------------------------------------------------===//
256 // Alias instruction that maps zero vector to pxor / xorp* for sse.
257 // This is expanded by ExpandPostRAPseudos to an xorps / vxorps, and then
258 // swizzled by ExecutionDepsFix to pxor.
259 // We set canFoldAsLoad because this can be converted to a constant-pool
260 // load of an all-zeros value if folding it would be beneficial.
261 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
262 isPseudo = 1, neverHasSideEffects = 1 in {
263 def V_SET0 : I<0, Pseudo, (outs VR128:$dst), (ins), "", []>;
266 def : Pat<(v4f32 immAllZerosV), (V_SET0)>;
267 def : Pat<(v2f64 immAllZerosV), (V_SET0)>;
268 def : Pat<(v4i32 immAllZerosV), (V_SET0)>;
269 def : Pat<(v2i64 immAllZerosV), (V_SET0)>;
270 def : Pat<(v8i16 immAllZerosV), (V_SET0)>;
271 def : Pat<(v16i8 immAllZerosV), (V_SET0)>;
274 // The same as done above but for AVX. The 256-bit ISA does not support PI,
275 // and doesn't need it because on sandy bridge the register is set to zero
276 // at the rename stage without using any execution unit, so SET0PSY
277 // and SET0PDY can be used for vector int instructions without penalty
278 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
279 // JIT implementatioan, it does not expand the instructions below like
280 // X86MCInstLower does.
281 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
282 isCodeGenOnly = 1, Predicates = [HasAVX] in {
283 def AVX_SET0PSY : PSI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
284 [(set VR256:$dst, (v8f32 immAllZerosV))]>, VEX_4V;
285 def AVX_SET0PDY : PDI<0x57, MRMInitReg, (outs VR256:$dst), (ins), "",
286 [(set VR256:$dst, (v4f64 immAllZerosV))]>, VEX_4V;
290 // AVX has no support for 256-bit integer instructions, but since the 128-bit
291 // VPXOR instruction writes zero to its upper part, it's safe build zeros.
292 def : Pat<(v8i32 immAllZerosV), (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
293 def : Pat<(bc_v8i32 (v8f32 immAllZerosV)),
294 (SUBREG_TO_REG (i32 0), (V_SET0), sub_xmm)>;
296 def : Pat<(v4i64 immAllZerosV), (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
297 def : Pat<(bc_v4i64 (v8f32 immAllZerosV)),
298 (SUBREG_TO_REG (i64 0), (V_SET0), sub_xmm)>;
300 // We set canFoldAsLoad because this can be converted to a constant-pool
301 // load of an all-ones value if folding it would be beneficial.
302 // FIXME: Change encoding to pseudo! This is blocked right now by the x86
303 // JIT implementation, it does not expand the instructions below like
304 // X86MCInstLower does.
305 let isReMaterializable = 1, isAsCheapAsAMove = 1, canFoldAsLoad = 1,
306 isCodeGenOnly = 1, ExeDomain = SSEPackedInt in {
307 def V_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
308 [(set VR128:$dst, (v4i32 immAllOnesV))]>;
309 let Predicates = [HasAVX] in
310 def AVX_SETALLONES : PDI<0x76, MRMInitReg, (outs VR128:$dst), (ins), "",
311 [(set VR128:$dst, (v4i32 immAllOnesV))]>, VEX_4V;
312 let Predicates = [HasAVX2] in
313 def AVX2_SETALLONES : PDI<0x76, MRMInitReg, (outs VR256:$dst), (ins), "",
314 [(set VR256:$dst, (v8i32 immAllOnesV))]>, VEX_4V;
318 //===----------------------------------------------------------------------===//
319 // SSE 1 & 2 - Move FP Scalar Instructions
321 // Move Instructions. Register-to-register movss/movsd is not used for FR32/64
322 // register copies because it's a partial register update; FsMOVAPSrr/FsMOVAPDrr
323 // is used instead. Register-to-register movss/movsd is not modeled as an
324 // INSERT_SUBREG because INSERT_SUBREG requires that the insert be implementable
325 // in terms of a copy, and just mentioned, we don't use movss/movsd for copies.
326 //===----------------------------------------------------------------------===//
328 class sse12_move_rr<RegisterClass RC, ValueType vt, string asm> :
329 SI<0x10, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src1, RC:$src2), asm,
330 [(set (vt VR128:$dst), (movl VR128:$src1, (scalar_to_vector RC:$src2)))]>;
332 // Loading from memory automatically zeroing upper bits.
333 class sse12_move_rm<RegisterClass RC, X86MemOperand x86memop,
334 PatFrag mem_pat, string OpcodeStr> :
335 SI<0x10, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
336 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
337 [(set RC:$dst, (mem_pat addr:$src))]>;
340 def VMOVSSrr : sse12_move_rr<FR32, v4f32,
341 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XS, VEX_4V,
343 def VMOVSDrr : sse12_move_rr<FR64, v2f64,
344 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}">, XD, VEX_4V,
347 // For the disassembler
348 let isCodeGenOnly = 1 in {
349 def VMOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
350 (ins VR128:$src1, FR32:$src2),
351 "movss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
353 def VMOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
354 (ins VR128:$src1, FR64:$src2),
355 "movsd\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
359 let canFoldAsLoad = 1, isReMaterializable = 1 in {
360 def VMOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS, VEX,
362 let AddedComplexity = 20 in
363 def VMOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD, VEX,
367 def VMOVSSmr : SI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
368 "movss\t{$src, $dst|$dst, $src}",
369 [(store FR32:$src, addr:$dst)]>, XS, VEX, VEX_LIG;
370 def VMOVSDmr : SI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
371 "movsd\t{$src, $dst|$dst, $src}",
372 [(store FR64:$src, addr:$dst)]>, XD, VEX, VEX_LIG;
375 let Constraints = "$src1 = $dst" in {
376 def MOVSSrr : sse12_move_rr<FR32, v4f32,
377 "movss\t{$src2, $dst|$dst, $src2}">, XS;
378 def MOVSDrr : sse12_move_rr<FR64, v2f64,
379 "movsd\t{$src2, $dst|$dst, $src2}">, XD;
381 // For the disassembler
382 let isCodeGenOnly = 1 in {
383 def MOVSSrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
384 (ins VR128:$src1, FR32:$src2),
385 "movss\t{$src2, $dst|$dst, $src2}", []>, XS;
386 def MOVSDrr_REV : SI<0x11, MRMDestReg, (outs VR128:$dst),
387 (ins VR128:$src1, FR64:$src2),
388 "movsd\t{$src2, $dst|$dst, $src2}", []>, XD;
392 let canFoldAsLoad = 1, isReMaterializable = 1 in {
393 def MOVSSrm : sse12_move_rm<FR32, f32mem, loadf32, "movss">, XS;
395 let AddedComplexity = 20 in
396 def MOVSDrm : sse12_move_rm<FR64, f64mem, loadf64, "movsd">, XD;
399 def MOVSSmr : SSI<0x11, MRMDestMem, (outs), (ins f32mem:$dst, FR32:$src),
400 "movss\t{$src, $dst|$dst, $src}",
401 [(store FR32:$src, addr:$dst)]>;
402 def MOVSDmr : SDI<0x11, MRMDestMem, (outs), (ins f64mem:$dst, FR64:$src),
403 "movsd\t{$src, $dst|$dst, $src}",
404 [(store FR64:$src, addr:$dst)]>;
407 let Predicates = [HasSSE1] in {
408 let AddedComplexity = 15 in {
409 // Extract the low 32-bit value from one vector and insert it into another.
410 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
411 (MOVSSrr (v4f32 VR128:$src1),
412 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
413 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
414 (MOVSSrr (v4i32 VR128:$src1),
415 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
417 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
418 // MOVSS to the lower bits.
419 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
420 (MOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
421 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
422 (MOVSSrr (v4f32 (V_SET0)),
423 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
424 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
425 (MOVSSrr (v4i32 (V_SET0)),
426 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
429 let AddedComplexity = 20 in {
430 // MOVSSrm zeros the high parts of the register; represent this
431 // with SUBREG_TO_REG.
432 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
433 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
434 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
435 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
436 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
437 (SUBREG_TO_REG (i32 0), (MOVSSrm addr:$src), sub_ss)>;
440 // Extract and store.
441 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
444 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
446 // Shuffle with MOVSS
447 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
448 (MOVSSrr VR128:$src1, FR32:$src2)>;
449 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
450 (MOVSSrr (v4i32 VR128:$src1),
451 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
452 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
453 (MOVSSrr (v4f32 VR128:$src1),
454 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
457 let Predicates = [HasSSE2] in {
458 let AddedComplexity = 15 in {
459 // Extract the low 64-bit value from one vector and insert it into another.
460 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
461 (MOVSDrr (v2f64 VR128:$src1),
462 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
463 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
464 (MOVSDrr (v2i64 VR128:$src1),
465 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
467 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
468 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
469 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
470 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
471 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
473 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
474 // MOVSD to the lower bits.
475 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
476 (MOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
479 let AddedComplexity = 20 in {
480 // MOVSDrm zeros the high parts of the register; represent this
481 // with SUBREG_TO_REG.
482 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
483 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
484 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
485 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
486 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
487 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
488 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
489 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
490 def : Pat<(v2f64 (X86vzload addr:$src)),
491 (SUBREG_TO_REG (i64 0), (MOVSDrm addr:$src), sub_sd)>;
494 // Extract and store.
495 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
498 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
500 // Shuffle with MOVSD
501 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
502 (MOVSDrr VR128:$src1, FR64:$src2)>;
503 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
504 (MOVSDrr (v2i64 VR128:$src1),
505 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
506 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
507 (MOVSDrr (v2f64 VR128:$src1),
508 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
509 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
510 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
511 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
512 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
514 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
515 // is during lowering, where it's not possible to recognize the fold cause
516 // it has two uses through a bitcast. One use disappears at isel time and the
517 // fold opportunity reappears.
518 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
519 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),sub_sd))>;
520 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
521 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),sub_sd))>;
522 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
523 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),sub_sd))>;
524 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
525 (MOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),sub_sd))>;
528 let Predicates = [HasAVX] in {
529 let AddedComplexity = 15 in {
530 // Extract the low 32-bit value from one vector and insert it into another.
531 def : Pat<(v4f32 (movl VR128:$src1, VR128:$src2)),
532 (VMOVSSrr (v4f32 VR128:$src1),
533 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
534 def : Pat<(v4i32 (movl VR128:$src1, VR128:$src2)),
535 (VMOVSSrr (v4i32 VR128:$src1),
536 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
538 // Extract the low 64-bit value from one vector and insert it into another.
539 def : Pat<(v2f64 (movl VR128:$src1, VR128:$src2)),
540 (VMOVSDrr (v2f64 VR128:$src1),
541 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
542 def : Pat<(v2i64 (movl VR128:$src1, VR128:$src2)),
543 (VMOVSDrr (v2i64 VR128:$src1),
544 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
546 // vector_shuffle v1, v2 <4, 5, 2, 3> using movsd
547 def : Pat<(v4f32 (movlp VR128:$src1, VR128:$src2)),
548 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
549 def : Pat<(v4i32 (movlp VR128:$src1, VR128:$src2)),
550 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG VR128:$src2, sub_sd))>;
552 // Move scalar to XMM zero-extended, zeroing a VR128 then do a
553 // MOVS{S,D} to the lower bits.
554 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector FR32:$src)))),
555 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)>;
556 def : Pat<(v4f32 (X86vzmovl (v4f32 VR128:$src))),
557 (VMOVSSrr (v4f32 (V_SET0)),
558 (f32 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)))>;
559 def : Pat<(v4i32 (X86vzmovl (v4i32 VR128:$src))),
560 (VMOVSSrr (v4i32 (V_SET0)),
561 (EXTRACT_SUBREG (v4i32 VR128:$src), sub_ss))>;
562 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector FR64:$src)))),
563 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)>;
565 // Move low f32 and clear high bits.
566 def : Pat<(v8f32 (X86vzmovl (v8f32 VR256:$src))),
567 (SUBREG_TO_REG (i32 0),
568 (VMOVSSrr (v4f32 (V_SET0)),
569 (EXTRACT_SUBREG (v8f32 VR256:$src), sub_ss)), sub_xmm)>;
570 def : Pat<(v8i32 (X86vzmovl (v8i32 VR256:$src))),
571 (SUBREG_TO_REG (i32 0),
572 (VMOVSSrr (v4i32 (V_SET0)),
573 (EXTRACT_SUBREG (v8i32 VR256:$src), sub_ss)), sub_xmm)>;
576 let AddedComplexity = 20 in {
577 // MOVSSrm zeros the high parts of the register; represent this
578 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
579 def : Pat<(v4f32 (X86vzmovl (v4f32 (scalar_to_vector (loadf32 addr:$src))))),
580 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
581 def : Pat<(v4f32 (scalar_to_vector (loadf32 addr:$src))),
582 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
583 def : Pat<(v4f32 (X86vzmovl (loadv4f32 addr:$src))),
584 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
586 // MOVSDrm zeros the high parts of the register; represent this
587 // with SUBREG_TO_REG. The AVX versions also write: DST[255:128] <- 0
588 def : Pat<(v2f64 (X86vzmovl (v2f64 (scalar_to_vector (loadf64 addr:$src))))),
589 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
590 def : Pat<(v2f64 (scalar_to_vector (loadf64 addr:$src))),
591 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
592 def : Pat<(v2f64 (X86vzmovl (loadv2f64 addr:$src))),
593 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
594 def : Pat<(v2f64 (X86vzmovl (bc_v2f64 (loadv4f32 addr:$src)))),
595 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
596 def : Pat<(v2f64 (X86vzload addr:$src)),
597 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
599 // Represent the same patterns above but in the form they appear for
601 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
602 (v4i32 (scalar_to_vector (loadi32 addr:$src))), (i32 0)))),
603 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
604 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
605 (v4f32 (scalar_to_vector (loadf32 addr:$src))), (i32 0)))),
606 (SUBREG_TO_REG (i32 0), (VMOVSSrm addr:$src), sub_ss)>;
607 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
608 (v2f64 (scalar_to_vector (loadf64 addr:$src))), (i32 0)))),
609 (SUBREG_TO_REG (i32 0), (VMOVSDrm addr:$src), sub_sd)>;
611 def : Pat<(v8f32 (X86vzmovl (insert_subvector undef,
612 (v4f32 (scalar_to_vector FR32:$src)), (i32 0)))),
613 (SUBREG_TO_REG (i32 0),
614 (v4f32 (VMOVSSrr (v4f32 (V_SET0)), FR32:$src)),
616 def : Pat<(v4f64 (X86vzmovl (insert_subvector undef,
617 (v2f64 (scalar_to_vector FR64:$src)), (i32 0)))),
618 (SUBREG_TO_REG (i64 0),
619 (v2f64 (VMOVSDrr (v2f64 (V_SET0)), FR64:$src)),
621 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
622 (v2i64 (scalar_to_vector (loadi64 addr:$src))), (i32 0)))),
623 (SUBREG_TO_REG (i64 0), (VMOVSDrm addr:$src), sub_sd)>;
625 // Move low f64 and clear high bits.
626 def : Pat<(v4f64 (X86vzmovl (v4f64 VR256:$src))),
627 (SUBREG_TO_REG (i32 0),
628 (VMOVSDrr (v2f64 (V_SET0)),
629 (EXTRACT_SUBREG (v4f64 VR256:$src), sub_sd)), sub_xmm)>;
631 def : Pat<(v4i64 (X86vzmovl (v4i64 VR256:$src))),
632 (SUBREG_TO_REG (i32 0),
633 (VMOVSDrr (v2i64 (V_SET0)),
634 (EXTRACT_SUBREG (v4i64 VR256:$src), sub_sd)), sub_xmm)>;
636 // Extract and store.
637 def : Pat<(store (f32 (vector_extract (v4f32 VR128:$src), (iPTR 0))),
640 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
641 def : Pat<(store (f64 (vector_extract (v2f64 VR128:$src), (iPTR 0))),
644 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd))>;
646 // Shuffle with VMOVSS
647 def : Pat<(v4f32 (X86Movss VR128:$src1, (scalar_to_vector FR32:$src2))),
648 (VMOVSSrr VR128:$src1, FR32:$src2)>;
649 def : Pat<(v4i32 (X86Movss VR128:$src1, VR128:$src2)),
650 (VMOVSSrr (v4i32 VR128:$src1),
651 (EXTRACT_SUBREG (v4i32 VR128:$src2), sub_ss))>;
652 def : Pat<(v4f32 (X86Movss VR128:$src1, VR128:$src2)),
653 (VMOVSSrr (v4f32 VR128:$src1),
654 (EXTRACT_SUBREG (v4f32 VR128:$src2), sub_ss))>;
657 def : Pat<(v8i32 (X86Movsd VR256:$src1, VR256:$src2)),
658 (SUBREG_TO_REG (i32 0),
659 (VMOVSSrr (EXTRACT_SUBREG (v8i32 VR256:$src1), sub_ss),
660 (EXTRACT_SUBREG (v8i32 VR256:$src2), sub_ss)), sub_xmm)>;
661 def : Pat<(v8f32 (X86Movsd VR256:$src1, VR256:$src2)),
662 (SUBREG_TO_REG (i32 0),
663 (VMOVSSrr (EXTRACT_SUBREG (v8f32 VR256:$src1), sub_ss),
664 (EXTRACT_SUBREG (v8f32 VR256:$src2), sub_ss)), sub_xmm)>;
666 // Shuffle with VMOVSD
667 def : Pat<(v2f64 (X86Movsd VR128:$src1, (scalar_to_vector FR64:$src2))),
668 (VMOVSDrr VR128:$src1, FR64:$src2)>;
669 def : Pat<(v2i64 (X86Movsd VR128:$src1, VR128:$src2)),
670 (VMOVSDrr (v2i64 VR128:$src1),
671 (EXTRACT_SUBREG (v2i64 VR128:$src2), sub_sd))>;
672 def : Pat<(v2f64 (X86Movsd VR128:$src1, VR128:$src2)),
673 (VMOVSDrr (v2f64 VR128:$src1),
674 (EXTRACT_SUBREG (v2f64 VR128:$src2), sub_sd))>;
675 def : Pat<(v4f32 (X86Movsd VR128:$src1, VR128:$src2)),
676 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
678 def : Pat<(v4i32 (X86Movsd VR128:$src1, VR128:$src2)),
679 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
683 def : Pat<(v4i64 (X86Movsd VR256:$src1, VR256:$src2)),
684 (SUBREG_TO_REG (i32 0),
685 (VMOVSDrr (EXTRACT_SUBREG (v4i64 VR256:$src1), sub_sd),
686 (EXTRACT_SUBREG (v4i64 VR256:$src2), sub_sd)), sub_xmm)>;
687 def : Pat<(v4f64 (X86Movsd VR256:$src1, VR256:$src2)),
688 (SUBREG_TO_REG (i32 0),
689 (VMOVSDrr (EXTRACT_SUBREG (v4f64 VR256:$src1), sub_sd),
690 (EXTRACT_SUBREG (v4f64 VR256:$src2), sub_sd)), sub_xmm)>;
693 // FIXME: Instead of a X86Movlps there should be a X86Movsd here, the problem
694 // is during lowering, where it's not possible to recognize the fold cause
695 // it has two uses through a bitcast. One use disappears at isel time and the
696 // fold opportunity reappears.
697 def : Pat<(v2f64 (X86Movlpd VR128:$src1, VR128:$src2)),
698 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2f64 VR128:$src2),
700 def : Pat<(v2i64 (X86Movlpd VR128:$src1, VR128:$src2)),
701 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v2i64 VR128:$src2),
703 def : Pat<(v4f32 (X86Movlps VR128:$src1, VR128:$src2)),
704 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4f32 VR128:$src2),
706 def : Pat<(v4i32 (X86Movlps VR128:$src1, VR128:$src2)),
707 (VMOVSDrr VR128:$src1, (EXTRACT_SUBREG (v4i32 VR128:$src2),
711 //===----------------------------------------------------------------------===//
712 // SSE 1 & 2 - Move Aligned/Unaligned FP Instructions
713 //===----------------------------------------------------------------------===//
715 multiclass sse12_mov_packed<bits<8> opc, RegisterClass RC,
716 X86MemOperand x86memop, PatFrag ld_frag,
717 string asm, Domain d,
718 bit IsReMaterializable = 1> {
719 let neverHasSideEffects = 1 in
720 def rr : PI<opc, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
721 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>;
722 let canFoldAsLoad = 1, isReMaterializable = IsReMaterializable in
723 def rm : PI<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
724 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
725 [(set RC:$dst, (ld_frag addr:$src))], d>;
728 defm VMOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
729 "movaps", SSEPackedSingle>, TB, VEX;
730 defm VMOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
731 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
732 defm VMOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
733 "movups", SSEPackedSingle>, TB, VEX;
734 defm VMOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
735 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
737 defm VMOVAPSY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv8f32,
738 "movaps", SSEPackedSingle>, TB, VEX;
739 defm VMOVAPDY : sse12_mov_packed<0x28, VR256, f256mem, alignedloadv4f64,
740 "movapd", SSEPackedDouble>, TB, OpSize, VEX;
741 defm VMOVUPSY : sse12_mov_packed<0x10, VR256, f256mem, loadv8f32,
742 "movups", SSEPackedSingle>, TB, VEX;
743 defm VMOVUPDY : sse12_mov_packed<0x10, VR256, f256mem, loadv4f64,
744 "movupd", SSEPackedDouble, 0>, TB, OpSize, VEX;
745 defm MOVAPS : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv4f32,
746 "movaps", SSEPackedSingle>, TB;
747 defm MOVAPD : sse12_mov_packed<0x28, VR128, f128mem, alignedloadv2f64,
748 "movapd", SSEPackedDouble>, TB, OpSize;
749 defm MOVUPS : sse12_mov_packed<0x10, VR128, f128mem, loadv4f32,
750 "movups", SSEPackedSingle>, TB;
751 defm MOVUPD : sse12_mov_packed<0x10, VR128, f128mem, loadv2f64,
752 "movupd", SSEPackedDouble, 0>, TB, OpSize;
754 def VMOVAPSmr : VPSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
755 "movaps\t{$src, $dst|$dst, $src}",
756 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>, VEX;
757 def VMOVAPDmr : VPDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
758 "movapd\t{$src, $dst|$dst, $src}",
759 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>, VEX;
760 def VMOVUPSmr : VPSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
761 "movups\t{$src, $dst|$dst, $src}",
762 [(store (v4f32 VR128:$src), addr:$dst)]>, VEX;
763 def VMOVUPDmr : VPDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
764 "movupd\t{$src, $dst|$dst, $src}",
765 [(store (v2f64 VR128:$src), addr:$dst)]>, VEX;
766 def VMOVAPSYmr : VPSI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
767 "movaps\t{$src, $dst|$dst, $src}",
768 [(alignedstore256 (v8f32 VR256:$src), addr:$dst)]>, VEX;
769 def VMOVAPDYmr : VPDI<0x29, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
770 "movapd\t{$src, $dst|$dst, $src}",
771 [(alignedstore256 (v4f64 VR256:$src), addr:$dst)]>, VEX;
772 def VMOVUPSYmr : VPSI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
773 "movups\t{$src, $dst|$dst, $src}",
774 [(store (v8f32 VR256:$src), addr:$dst)]>, VEX;
775 def VMOVUPDYmr : VPDI<0x11, MRMDestMem, (outs), (ins f256mem:$dst, VR256:$src),
776 "movupd\t{$src, $dst|$dst, $src}",
777 [(store (v4f64 VR256:$src), addr:$dst)]>, VEX;
780 let isCodeGenOnly = 1 in {
781 def VMOVAPSrr_REV : VPSI<0x29, MRMDestReg, (outs VR128:$dst),
783 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
784 def VMOVAPDrr_REV : VPDI<0x29, MRMDestReg, (outs VR128:$dst),
786 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
787 def VMOVUPSrr_REV : VPSI<0x11, MRMDestReg, (outs VR128:$dst),
789 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
790 def VMOVUPDrr_REV : VPDI<0x11, MRMDestReg, (outs VR128:$dst),
792 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
793 def VMOVAPSYrr_REV : VPSI<0x29, MRMDestReg, (outs VR256:$dst),
795 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
796 def VMOVAPDYrr_REV : VPDI<0x29, MRMDestReg, (outs VR256:$dst),
798 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
799 def VMOVUPSYrr_REV : VPSI<0x11, MRMDestReg, (outs VR256:$dst),
801 "movups\t{$src, $dst|$dst, $src}", []>, VEX;
802 def VMOVUPDYrr_REV : VPDI<0x11, MRMDestReg, (outs VR256:$dst),
804 "movupd\t{$src, $dst|$dst, $src}", []>, VEX;
807 let Predicates = [HasAVX] in {
808 def : Pat<(v8i32 (X86vzmovl
809 (insert_subvector undef, (v4i32 VR128:$src), (i32 0)))),
810 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
811 def : Pat<(v4i64 (X86vzmovl
812 (insert_subvector undef, (v2i64 VR128:$src), (i32 0)))),
813 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
814 def : Pat<(v8f32 (X86vzmovl
815 (insert_subvector undef, (v4f32 VR128:$src), (i32 0)))),
816 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
817 def : Pat<(v4f64 (X86vzmovl
818 (insert_subvector undef, (v2f64 VR128:$src), (i32 0)))),
819 (SUBREG_TO_REG (i32 0), (VMOVAPSrr VR128:$src), sub_xmm)>;
823 def : Pat<(int_x86_avx_loadu_ps_256 addr:$src), (VMOVUPSYrm addr:$src)>;
824 def : Pat<(int_x86_avx_storeu_ps_256 addr:$dst, VR256:$src),
825 (VMOVUPSYmr addr:$dst, VR256:$src)>;
827 def : Pat<(int_x86_avx_loadu_pd_256 addr:$src), (VMOVUPDYrm addr:$src)>;
828 def : Pat<(int_x86_avx_storeu_pd_256 addr:$dst, VR256:$src),
829 (VMOVUPDYmr addr:$dst, VR256:$src)>;
831 def MOVAPSmr : PSI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
832 "movaps\t{$src, $dst|$dst, $src}",
833 [(alignedstore (v4f32 VR128:$src), addr:$dst)]>;
834 def MOVAPDmr : PDI<0x29, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
835 "movapd\t{$src, $dst|$dst, $src}",
836 [(alignedstore (v2f64 VR128:$src), addr:$dst)]>;
837 def MOVUPSmr : PSI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
838 "movups\t{$src, $dst|$dst, $src}",
839 [(store (v4f32 VR128:$src), addr:$dst)]>;
840 def MOVUPDmr : PDI<0x11, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
841 "movupd\t{$src, $dst|$dst, $src}",
842 [(store (v2f64 VR128:$src), addr:$dst)]>;
845 let isCodeGenOnly = 1 in {
846 def MOVAPSrr_REV : PSI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
847 "movaps\t{$src, $dst|$dst, $src}", []>;
848 def MOVAPDrr_REV : PDI<0x29, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
849 "movapd\t{$src, $dst|$dst, $src}", []>;
850 def MOVUPSrr_REV : PSI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
851 "movups\t{$src, $dst|$dst, $src}", []>;
852 def MOVUPDrr_REV : PDI<0x11, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
853 "movupd\t{$src, $dst|$dst, $src}", []>;
856 let Predicates = [HasAVX] in {
857 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
858 (VMOVUPSmr addr:$dst, VR128:$src)>;
859 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
860 (VMOVUPDmr addr:$dst, VR128:$src)>;
863 let Predicates = [HasSSE1] in
864 def : Pat<(int_x86_sse_storeu_ps addr:$dst, VR128:$src),
865 (MOVUPSmr addr:$dst, VR128:$src)>;
866 let Predicates = [HasSSE2] in
867 def : Pat<(int_x86_sse2_storeu_pd addr:$dst, VR128:$src),
868 (MOVUPDmr addr:$dst, VR128:$src)>;
870 // Use movaps / movups for SSE integer load / store (one byte shorter).
871 // The instructions selected below are then converted to MOVDQA/MOVDQU
872 // during the SSE domain pass.
873 let Predicates = [HasSSE1] in {
874 def : Pat<(alignedloadv4i32 addr:$src),
875 (MOVAPSrm addr:$src)>;
876 def : Pat<(loadv4i32 addr:$src),
877 (MOVUPSrm addr:$src)>;
878 def : Pat<(alignedloadv2i64 addr:$src),
879 (MOVAPSrm addr:$src)>;
880 def : Pat<(loadv2i64 addr:$src),
881 (MOVUPSrm addr:$src)>;
883 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
884 (MOVAPSmr addr:$dst, VR128:$src)>;
885 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
886 (MOVAPSmr addr:$dst, VR128:$src)>;
887 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
888 (MOVAPSmr addr:$dst, VR128:$src)>;
889 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
890 (MOVAPSmr addr:$dst, VR128:$src)>;
891 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
892 (MOVUPSmr addr:$dst, VR128:$src)>;
893 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
894 (MOVUPSmr addr:$dst, VR128:$src)>;
895 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
896 (MOVUPSmr addr:$dst, VR128:$src)>;
897 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
898 (MOVUPSmr addr:$dst, VR128:$src)>;
901 // Use vmovaps/vmovups for AVX integer load/store.
902 let Predicates = [HasAVX] in {
903 // 128-bit load/store
904 def : Pat<(alignedloadv4i32 addr:$src),
905 (VMOVAPSrm addr:$src)>;
906 def : Pat<(loadv4i32 addr:$src),
907 (VMOVUPSrm addr:$src)>;
908 def : Pat<(alignedloadv2i64 addr:$src),
909 (VMOVAPSrm addr:$src)>;
910 def : Pat<(loadv2i64 addr:$src),
911 (VMOVUPSrm addr:$src)>;
913 def : Pat<(alignedstore (v2i64 VR128:$src), addr:$dst),
914 (VMOVAPSmr addr:$dst, VR128:$src)>;
915 def : Pat<(alignedstore (v4i32 VR128:$src), addr:$dst),
916 (VMOVAPSmr addr:$dst, VR128:$src)>;
917 def : Pat<(alignedstore (v8i16 VR128:$src), addr:$dst),
918 (VMOVAPSmr addr:$dst, VR128:$src)>;
919 def : Pat<(alignedstore (v16i8 VR128:$src), addr:$dst),
920 (VMOVAPSmr addr:$dst, VR128:$src)>;
921 def : Pat<(store (v2i64 VR128:$src), addr:$dst),
922 (VMOVUPSmr addr:$dst, VR128:$src)>;
923 def : Pat<(store (v4i32 VR128:$src), addr:$dst),
924 (VMOVUPSmr addr:$dst, VR128:$src)>;
925 def : Pat<(store (v8i16 VR128:$src), addr:$dst),
926 (VMOVUPSmr addr:$dst, VR128:$src)>;
927 def : Pat<(store (v16i8 VR128:$src), addr:$dst),
928 (VMOVUPSmr addr:$dst, VR128:$src)>;
930 // 256-bit load/store
931 def : Pat<(alignedloadv4i64 addr:$src),
932 (VMOVAPSYrm addr:$src)>;
933 def : Pat<(loadv4i64 addr:$src),
934 (VMOVUPSYrm addr:$src)>;
935 def : Pat<(alignedloadv8i32 addr:$src),
936 (VMOVAPSYrm addr:$src)>;
937 def : Pat<(loadv8i32 addr:$src),
938 (VMOVUPSYrm addr:$src)>;
939 def : Pat<(alignedstore256 (v4i64 VR256:$src), addr:$dst),
940 (VMOVAPSYmr addr:$dst, VR256:$src)>;
941 def : Pat<(alignedstore256 (v8i32 VR256:$src), addr:$dst),
942 (VMOVAPSYmr addr:$dst, VR256:$src)>;
943 def : Pat<(alignedstore256 (v16i16 VR256:$src), addr:$dst),
944 (VMOVAPSYmr addr:$dst, VR256:$src)>;
945 def : Pat<(alignedstore256 (v32i8 VR256:$src), addr:$dst),
946 (VMOVAPSYmr addr:$dst, VR256:$src)>;
947 def : Pat<(store (v4i64 VR256:$src), addr:$dst),
948 (VMOVUPSYmr addr:$dst, VR256:$src)>;
949 def : Pat<(store (v8i32 VR256:$src), addr:$dst),
950 (VMOVUPSYmr addr:$dst, VR256:$src)>;
951 def : Pat<(store (v16i16 VR256:$src), addr:$dst),
952 (VMOVUPSYmr addr:$dst, VR256:$src)>;
953 def : Pat<(store (v32i8 VR256:$src), addr:$dst),
954 (VMOVUPSYmr addr:$dst, VR256:$src)>;
957 // Alias instruction to do FR32 or FR64 reg-to-reg copy using movaps. Upper
958 // bits are disregarded. FIXME: Set encoding to pseudo!
959 let neverHasSideEffects = 1 in {
960 def FsMOVAPSrr : PSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
961 "movaps\t{$src, $dst|$dst, $src}", []>;
962 def FsMOVAPDrr : PDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
963 "movapd\t{$src, $dst|$dst, $src}", []>;
964 def FsVMOVAPSrr : VPSI<0x28, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
965 "movaps\t{$src, $dst|$dst, $src}", []>, VEX;
966 def FsVMOVAPDrr : VPDI<0x28, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
967 "movapd\t{$src, $dst|$dst, $src}", []>, VEX;
970 // Alias instruction to load FR32 or FR64 from f128mem using movaps. Upper
971 // bits are disregarded. FIXME: Set encoding to pseudo!
972 let canFoldAsLoad = 1, isReMaterializable = 1 in {
973 def FsMOVAPSrm : PSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
974 "movaps\t{$src, $dst|$dst, $src}",
975 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>;
976 def FsMOVAPDrm : PDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
977 "movapd\t{$src, $dst|$dst, $src}",
978 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>;
979 let isCodeGenOnly = 1 in {
980 def FsVMOVAPSrm : VPSI<0x28, MRMSrcMem, (outs FR32:$dst), (ins f128mem:$src),
981 "movaps\t{$src, $dst|$dst, $src}",
982 [(set FR32:$dst, (alignedloadfsf32 addr:$src))]>, VEX;
983 def FsVMOVAPDrm : VPDI<0x28, MRMSrcMem, (outs FR64:$dst), (ins f128mem:$src),
984 "movapd\t{$src, $dst|$dst, $src}",
985 [(set FR64:$dst, (alignedloadfsf64 addr:$src))]>, VEX;
989 //===----------------------------------------------------------------------===//
990 // SSE 1 & 2 - Move Low packed FP Instructions
991 //===----------------------------------------------------------------------===//
993 multiclass sse12_mov_hilo_packed<bits<8>opc, RegisterClass RC,
994 PatFrag mov_frag, string base_opc,
996 def PSrm : PI<opc, MRMSrcMem,
997 (outs VR128:$dst), (ins VR128:$src1, f64mem:$src2),
998 !strconcat(base_opc, "s", asm_opr),
1001 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))))],
1002 SSEPackedSingle>, TB;
1004 def PDrm : PI<opc, MRMSrcMem,
1005 (outs RC:$dst), (ins RC:$src1, f64mem:$src2),
1006 !strconcat(base_opc, "d", asm_opr),
1007 [(set RC:$dst, (v2f64 (mov_frag RC:$src1,
1008 (scalar_to_vector (loadf64 addr:$src2)))))],
1009 SSEPackedDouble>, TB, OpSize;
1012 let AddedComplexity = 20 in {
1013 defm VMOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1014 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1016 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1017 defm MOVL : sse12_mov_hilo_packed<0x12, VR128, movlp, "movlp",
1018 "\t{$src2, $dst|$dst, $src2}">;
1021 def VMOVLPSmr : VPSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1022 "movlps\t{$src, $dst|$dst, $src}",
1023 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1024 (iPTR 0))), addr:$dst)]>, VEX;
1025 def VMOVLPDmr : VPDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1026 "movlpd\t{$src, $dst|$dst, $src}",
1027 [(store (f64 (vector_extract (v2f64 VR128:$src),
1028 (iPTR 0))), addr:$dst)]>, VEX;
1029 def MOVLPSmr : PSI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1030 "movlps\t{$src, $dst|$dst, $src}",
1031 [(store (f64 (vector_extract (bc_v2f64 (v4f32 VR128:$src)),
1032 (iPTR 0))), addr:$dst)]>;
1033 def MOVLPDmr : PDI<0x13, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1034 "movlpd\t{$src, $dst|$dst, $src}",
1035 [(store (f64 (vector_extract (v2f64 VR128:$src),
1036 (iPTR 0))), addr:$dst)]>;
1038 let Predicates = [HasAVX] in {
1039 let AddedComplexity = 20 in {
1040 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1041 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1042 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1043 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1044 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1045 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1046 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1047 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1048 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1049 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1052 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1053 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1054 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1055 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1056 VR128:$src2)), addr:$src1),
1057 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1059 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1060 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1061 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1062 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1063 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1065 // Shuffle with VMOVLPS
1066 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1067 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1068 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1069 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1070 def : Pat<(X86Movlps VR128:$src1,
1071 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1072 (VMOVLPSrm VR128:$src1, addr:$src2)>;
1074 // Shuffle with VMOVLPD
1075 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1076 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1077 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1078 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1079 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1080 (scalar_to_vector (loadf64 addr:$src2)))),
1081 (VMOVLPDrm VR128:$src1, addr:$src2)>;
1084 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1086 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1087 def : Pat<(store (v4i32 (X86Movlps
1088 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)), addr:$src1),
1089 (VMOVLPSmr addr:$src1, VR128:$src2)>;
1090 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1092 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1093 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1095 (VMOVLPDmr addr:$src1, VR128:$src2)>;
1098 let Predicates = [HasSSE1] in {
1099 let AddedComplexity = 20 in {
1100 // vector_shuffle v1, (load v2) <4, 5, 2, 3> using MOVLPS
1101 def : Pat<(v4f32 (movlp VR128:$src1, (load addr:$src2))),
1102 (MOVLPSrm VR128:$src1, addr:$src2)>;
1103 def : Pat<(v4i32 (movlp VR128:$src1, (load addr:$src2))),
1104 (MOVLPSrm VR128:$src1, addr:$src2)>;
1107 // (store (vector_shuffle (load addr), v2, <4, 5, 2, 3>), addr) using MOVLPS
1108 def : Pat<(store (i64 (vector_extract (bc_v2i64 (v4f32 VR128:$src2)),
1109 (iPTR 0))), addr:$src1),
1110 (MOVLPSmr addr:$src1, VR128:$src2)>;
1111 def : Pat<(store (v4f32 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1112 (MOVLPSmr addr:$src1, VR128:$src2)>;
1113 def : Pat<(store (v4i32 (movlp (bc_v4i32 (loadv2i64 addr:$src1)),
1114 VR128:$src2)), addr:$src1),
1115 (MOVLPSmr addr:$src1, VR128:$src2)>;
1117 // Shuffle with MOVLPS
1118 def : Pat<(v4f32 (X86Movlps VR128:$src1, (load addr:$src2))),
1119 (MOVLPSrm VR128:$src1, addr:$src2)>;
1120 def : Pat<(v4i32 (X86Movlps VR128:$src1, (load addr:$src2))),
1121 (MOVLPSrm VR128:$src1, addr:$src2)>;
1122 def : Pat<(X86Movlps VR128:$src1,
1123 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1124 (MOVLPSrm VR128:$src1, addr:$src2)>;
1125 def : Pat<(X86Movlps VR128:$src1,
1126 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1127 (MOVLPSrm VR128:$src1, addr:$src2)>;
1130 def : Pat<(store (v4f32 (X86Movlps (load addr:$src1), VR128:$src2)),
1132 (MOVLPSmr addr:$src1, VR128:$src2)>;
1133 def : Pat<(store (v4i32 (X86Movlps
1134 (bc_v4i32 (loadv2i64 addr:$src1)), VR128:$src2)),
1136 (MOVLPSmr addr:$src1, VR128:$src2)>;
1139 let Predicates = [HasSSE2] in {
1140 let AddedComplexity = 20 in {
1141 // vector_shuffle v1, (load v2) <2, 1> using MOVLPS
1142 def : Pat<(v2f64 (movlp VR128:$src1, (load addr:$src2))),
1143 (MOVLPDrm VR128:$src1, addr:$src2)>;
1144 def : Pat<(v2i64 (movlp VR128:$src1, (load addr:$src2))),
1145 (MOVLPDrm VR128:$src1, addr:$src2)>;
1148 // (store (vector_shuffle (load addr), v2, <2, 1>), addr) using MOVLPS
1149 def : Pat<(store (v2f64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1150 (MOVLPDmr addr:$src1, VR128:$src2)>;
1151 def : Pat<(store (v2i64 (movlp (load addr:$src1), VR128:$src2)), addr:$src1),
1152 (MOVLPDmr addr:$src1, VR128:$src2)>;
1154 // Shuffle with MOVLPD
1155 def : Pat<(v2f64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1156 (MOVLPDrm VR128:$src1, addr:$src2)>;
1157 def : Pat<(v2i64 (X86Movlpd VR128:$src1, (load addr:$src2))),
1158 (MOVLPDrm VR128:$src1, addr:$src2)>;
1159 def : Pat<(v2f64 (X86Movlpd VR128:$src1,
1160 (scalar_to_vector (loadf64 addr:$src2)))),
1161 (MOVLPDrm VR128:$src1, addr:$src2)>;
1164 def : Pat<(store (v2f64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1166 (MOVLPDmr addr:$src1, VR128:$src2)>;
1167 def : Pat<(store (v2i64 (X86Movlpd (load addr:$src1), VR128:$src2)),
1169 (MOVLPDmr addr:$src1, VR128:$src2)>;
1172 //===----------------------------------------------------------------------===//
1173 // SSE 1 & 2 - Move Hi packed FP Instructions
1174 //===----------------------------------------------------------------------===//
1176 let AddedComplexity = 20 in {
1177 defm VMOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1178 "\t{$src2, $src1, $dst|$dst, $src1, $src2}">, VEX_4V;
1180 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1181 defm MOVH : sse12_mov_hilo_packed<0x16, VR128, movlhps, "movhp",
1182 "\t{$src2, $dst|$dst, $src2}">;
1185 // v2f64 extract element 1 is always custom lowered to unpack high to low
1186 // and extract element 0 so the non-store version isn't too horrible.
1187 def VMOVHPSmr : VPSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1188 "movhps\t{$src, $dst|$dst, $src}",
1189 [(store (f64 (vector_extract
1190 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1191 (undef)), (iPTR 0))), addr:$dst)]>,
1193 def VMOVHPDmr : VPDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1194 "movhpd\t{$src, $dst|$dst, $src}",
1195 [(store (f64 (vector_extract
1196 (v2f64 (unpckh VR128:$src, (undef))),
1197 (iPTR 0))), addr:$dst)]>,
1199 def MOVHPSmr : PSI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1200 "movhps\t{$src, $dst|$dst, $src}",
1201 [(store (f64 (vector_extract
1202 (unpckh (bc_v2f64 (v4f32 VR128:$src)),
1203 (undef)), (iPTR 0))), addr:$dst)]>;
1204 def MOVHPDmr : PDI<0x17, MRMDestMem, (outs), (ins f64mem:$dst, VR128:$src),
1205 "movhpd\t{$src, $dst|$dst, $src}",
1206 [(store (f64 (vector_extract
1207 (v2f64 (unpckh VR128:$src, (undef))),
1208 (iPTR 0))), addr:$dst)]>;
1210 let Predicates = [HasAVX] in {
1212 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1213 (VMOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1214 def : Pat<(X86Movlhps VR128:$src1,
1215 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1216 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1217 def : Pat<(X86Movlhps VR128:$src1,
1218 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1219 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1220 def : Pat<(X86Movlhps VR128:$src1,
1221 (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1222 (VMOVHPSrm VR128:$src1, addr:$src2)>;
1224 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1225 // is during lowering, where it's not possible to recognize the load fold
1226 // cause it has two uses through a bitcast. One use disappears at isel time
1227 // and the fold opportunity reappears.
1228 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1229 (scalar_to_vector (loadf64 addr:$src2)))),
1230 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1232 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1233 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1234 (scalar_to_vector (loadf64 addr:$src2)))),
1235 (VMOVHPDrm VR128:$src1, addr:$src2)>;
1238 def : Pat<(store (f64 (vector_extract
1239 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1240 (VMOVHPSmr addr:$dst, VR128:$src)>;
1241 def : Pat<(store (f64 (vector_extract
1242 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1243 (VMOVHPDmr addr:$dst, VR128:$src)>;
1246 let Predicates = [HasSSE1] in {
1248 def : Pat<(movlhps VR128:$src1, (bc_v4i32 (v2i64 (X86vzload addr:$src2)))),
1249 (MOVHPSrm (v4i32 VR128:$src1), addr:$src2)>;
1250 def : Pat<(X86Movlhps VR128:$src1,
1251 (bc_v4f32 (v2f64 (scalar_to_vector (loadf64 addr:$src2))))),
1252 (MOVHPSrm VR128:$src1, addr:$src2)>;
1253 def : Pat<(X86Movlhps VR128:$src1,
1254 (bc_v4f32 (v2i64 (scalar_to_vector (loadi64 addr:$src2))))),
1255 (MOVHPSrm VR128:$src1, addr:$src2)>;
1256 def : Pat<(X86Movlhps VR128:$src1,
1257 (bc_v4f32 (v2i64 (X86vzload addr:$src2)))),
1258 (MOVHPSrm VR128:$src1, addr:$src2)>;
1261 def : Pat<(store (f64 (vector_extract
1262 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))), addr:$dst),
1263 (MOVHPSmr addr:$dst, VR128:$src)>;
1266 let Predicates = [HasSSE2] in {
1267 // FIXME: Instead of X86Unpckl, there should be a X86Movlhpd here, the problem
1268 // is during lowering, where it's not possible to recognize the load fold
1269 // cause it has two uses through a bitcast. One use disappears at isel time
1270 // and the fold opportunity reappears.
1271 def : Pat<(v2f64 (X86Unpckl VR128:$src1,
1272 (scalar_to_vector (loadf64 addr:$src2)))),
1273 (MOVHPDrm VR128:$src1, addr:$src2)>;
1275 // FIXME: This should be matched by a X86Movhpd instead. Same as above
1276 def : Pat<(v2f64 (X86Movlhpd VR128:$src1,
1277 (scalar_to_vector (loadf64 addr:$src2)))),
1278 (MOVHPDrm VR128:$src1, addr:$src2)>;
1281 def : Pat<(store (f64 (vector_extract
1282 (v2f64 (X86Unpckh VR128:$src, (undef))), (iPTR 0))),addr:$dst),
1283 (MOVHPDmr addr:$dst, VR128:$src)>;
1286 //===----------------------------------------------------------------------===//
1287 // SSE 1 & 2 - Move Low to High and High to Low packed FP Instructions
1288 //===----------------------------------------------------------------------===//
1290 let AddedComplexity = 20 in {
1291 def VMOVLHPSrr : VPSI<0x16, MRMSrcReg, (outs VR128:$dst),
1292 (ins VR128:$src1, VR128:$src2),
1293 "movlhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1295 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>,
1297 def VMOVHLPSrr : VPSI<0x12, MRMSrcReg, (outs VR128:$dst),
1298 (ins VR128:$src1, VR128:$src2),
1299 "movhlps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1301 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>,
1304 let Constraints = "$src1 = $dst", AddedComplexity = 20 in {
1305 def MOVLHPSrr : PSI<0x16, MRMSrcReg, (outs VR128:$dst),
1306 (ins VR128:$src1, VR128:$src2),
1307 "movlhps\t{$src2, $dst|$dst, $src2}",
1309 (v4f32 (movlhps VR128:$src1, VR128:$src2)))]>;
1310 def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
1311 (ins VR128:$src1, VR128:$src2),
1312 "movhlps\t{$src2, $dst|$dst, $src2}",
1314 (v4f32 (movhlps VR128:$src1, VR128:$src2)))]>;
1317 let Predicates = [HasAVX] in {
1319 let AddedComplexity = 20 in {
1320 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1321 (VMOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1322 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1323 (VMOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1325 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1326 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1327 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1329 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1330 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1331 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1332 (VMOVLHPSrr VR128:$src1, VR128:$src2)>;
1333 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1334 (VMOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1337 let AddedComplexity = 20 in {
1338 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1339 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1340 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1342 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1343 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1344 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1345 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1346 (VMOVHLPSrr VR128:$src1, VR128:$src1)>;
1349 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1350 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1351 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1352 (VMOVHLPSrr VR128:$src1, VR128:$src2)>;
1355 let Predicates = [HasSSE1] in {
1357 let AddedComplexity = 20 in {
1358 def : Pat<(v4f32 (movddup VR128:$src, (undef))),
1359 (MOVLHPSrr (v4f32 VR128:$src), (v4f32 VR128:$src))>;
1360 def : Pat<(v2i64 (movddup VR128:$src, (undef))),
1361 (MOVLHPSrr (v2i64 VR128:$src), (v2i64 VR128:$src))>;
1363 // vector_shuffle v1, v2 <0, 1, 4, 5> using MOVLHPS
1364 def : Pat<(v4i32 (movlhps VR128:$src1, VR128:$src2)),
1365 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1367 def : Pat<(v4f32 (X86Movlhps VR128:$src1, VR128:$src2)),
1368 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1369 def : Pat<(v4i32 (X86Movlhps VR128:$src1, VR128:$src2)),
1370 (MOVLHPSrr VR128:$src1, VR128:$src2)>;
1371 def : Pat<(v2i64 (X86Movlhps VR128:$src1, VR128:$src2)),
1372 (MOVLHPSrr (v2i64 VR128:$src1), VR128:$src2)>;
1375 let AddedComplexity = 20 in {
1376 // vector_shuffle v1, v2 <6, 7, 2, 3> using MOVHLPS
1377 def : Pat<(v4i32 (movhlps VR128:$src1, VR128:$src2)),
1378 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1380 // vector_shuffle v1, undef <2, ?, ?, ?> using MOVHLPS
1381 def : Pat<(v4f32 (movhlps_undef VR128:$src1, (undef))),
1382 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1383 def : Pat<(v4i32 (movhlps_undef VR128:$src1, (undef))),
1384 (MOVHLPSrr VR128:$src1, VR128:$src1)>;
1387 def : Pat<(v4f32 (X86Movhlps VR128:$src1, VR128:$src2)),
1388 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1389 def : Pat<(v4i32 (X86Movhlps VR128:$src1, VR128:$src2)),
1390 (MOVHLPSrr VR128:$src1, VR128:$src2)>;
1393 //===----------------------------------------------------------------------===//
1394 // SSE 1 & 2 - Conversion Instructions
1395 //===----------------------------------------------------------------------===//
1397 multiclass sse12_cvt_s<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1398 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1400 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1401 [(set DstRC:$dst, (OpNode SrcRC:$src))]>;
1402 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1403 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))]>;
1406 multiclass sse12_cvt_s_np<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1407 X86MemOperand x86memop, string asm> {
1408 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm, []>;
1410 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm, []>;
1413 multiclass sse12_cvt_p<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1414 SDNode OpNode, X86MemOperand x86memop, PatFrag ld_frag,
1415 string asm, Domain d> {
1416 def rr : PI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src), asm,
1417 [(set DstRC:$dst, (OpNode SrcRC:$src))], d>;
1418 def rm : PI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src), asm,
1419 [(set DstRC:$dst, (OpNode (ld_frag addr:$src)))], d>;
1422 multiclass sse12_vcvt_avx<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1423 X86MemOperand x86memop, string asm> {
1424 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src),
1425 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1427 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1428 (ins DstRC:$src1, x86memop:$src),
1429 !strconcat(asm,"\t{$src, $src1, $dst|$dst, $src1, $src}"), []>;
1432 defm VCVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1433 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1435 defm VCVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1436 "cvttss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1438 defm VCVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1439 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD, VEX,
1441 defm VCVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1442 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD,
1443 VEX, VEX_W, VEX_LIG;
1445 // The assembler can recognize rr 64-bit instructions by seeing a rxx
1446 // register, but the same isn't true when only using memory operands,
1447 // provide other assembly "l" and "q" forms to address this explicitly
1448 // where appropriate to do so.
1449 defm VCVTSI2SS : sse12_vcvt_avx<0x2A, GR32, FR32, i32mem, "cvtsi2ss">, XS,
1451 defm VCVTSI2SS64 : sse12_vcvt_avx<0x2A, GR64, FR32, i64mem, "cvtsi2ss{q}">, XS,
1452 VEX_4V, VEX_W, VEX_LIG;
1453 defm VCVTSI2SD : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd">, XD,
1455 defm VCVTSI2SDL : sse12_vcvt_avx<0x2A, GR32, FR64, i32mem, "cvtsi2sd{l}">, XD,
1457 defm VCVTSI2SD64 : sse12_vcvt_avx<0x2A, GR64, FR64, i64mem, "cvtsi2sd{q}">, XD,
1458 VEX_4V, VEX_W, VEX_LIG;
1460 let Predicates = [HasAVX] in {
1461 def : Pat<(f32 (sint_to_fp (loadi32 addr:$src))),
1462 (VCVTSI2SSrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1463 def : Pat<(f32 (sint_to_fp (loadi64 addr:$src))),
1464 (VCVTSI2SS64rm (f32 (IMPLICIT_DEF)), addr:$src)>;
1465 def : Pat<(f64 (sint_to_fp (loadi32 addr:$src))),
1466 (VCVTSI2SDrm (f64 (IMPLICIT_DEF)), addr:$src)>;
1467 def : Pat<(f64 (sint_to_fp (loadi64 addr:$src))),
1468 (VCVTSI2SD64rm (f64 (IMPLICIT_DEF)), addr:$src)>;
1470 def : Pat<(f32 (sint_to_fp GR32:$src)),
1471 (VCVTSI2SSrr (f32 (IMPLICIT_DEF)), GR32:$src)>;
1472 def : Pat<(f32 (sint_to_fp GR64:$src)),
1473 (VCVTSI2SS64rr (f32 (IMPLICIT_DEF)), GR64:$src)>;
1474 def : Pat<(f64 (sint_to_fp GR32:$src)),
1475 (VCVTSI2SDrr (f64 (IMPLICIT_DEF)), GR32:$src)>;
1476 def : Pat<(f64 (sint_to_fp GR64:$src)),
1477 (VCVTSI2SD64rr (f64 (IMPLICIT_DEF)), GR64:$src)>;
1480 defm CVTTSS2SI : sse12_cvt_s<0x2C, FR32, GR32, fp_to_sint, f32mem, loadf32,
1481 "cvttss2si\t{$src, $dst|$dst, $src}">, XS;
1482 defm CVTTSS2SI64 : sse12_cvt_s<0x2C, FR32, GR64, fp_to_sint, f32mem, loadf32,
1483 "cvttss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1484 defm CVTTSD2SI : sse12_cvt_s<0x2C, FR64, GR32, fp_to_sint, f64mem, loadf64,
1485 "cvttsd2si\t{$src, $dst|$dst, $src}">, XD;
1486 defm CVTTSD2SI64 : sse12_cvt_s<0x2C, FR64, GR64, fp_to_sint, f64mem, loadf64,
1487 "cvttsd2si{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1488 defm CVTSI2SS : sse12_cvt_s<0x2A, GR32, FR32, sint_to_fp, i32mem, loadi32,
1489 "cvtsi2ss\t{$src, $dst|$dst, $src}">, XS;
1490 defm CVTSI2SS64 : sse12_cvt_s<0x2A, GR64, FR32, sint_to_fp, i64mem, loadi64,
1491 "cvtsi2ss{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1492 defm CVTSI2SD : sse12_cvt_s<0x2A, GR32, FR64, sint_to_fp, i32mem, loadi32,
1493 "cvtsi2sd\t{$src, $dst|$dst, $src}">, XD;
1494 defm CVTSI2SD64 : sse12_cvt_s<0x2A, GR64, FR64, sint_to_fp, i64mem, loadi64,
1495 "cvtsi2sd{q}\t{$src, $dst|$dst, $src}">, XD, REX_W;
1497 // Conversion Instructions Intrinsics - Match intrinsics which expect MM
1498 // and/or XMM operand(s).
1500 multiclass sse12_cvt_sint<bits<8> opc, RegisterClass SrcRC, RegisterClass DstRC,
1501 Intrinsic Int, X86MemOperand x86memop, PatFrag ld_frag,
1503 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins SrcRC:$src),
1504 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1505 [(set DstRC:$dst, (Int SrcRC:$src))]>;
1506 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst), (ins x86memop:$src),
1507 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
1508 [(set DstRC:$dst, (Int (ld_frag addr:$src)))]>;
1511 multiclass sse12_cvt_sint_3addr<bits<8> opc, RegisterClass SrcRC,
1512 RegisterClass DstRC, Intrinsic Int, X86MemOperand x86memop,
1513 PatFrag ld_frag, string asm, bit Is2Addr = 1> {
1514 def rr : SI<opc, MRMSrcReg, (outs DstRC:$dst), (ins DstRC:$src1, SrcRC:$src2),
1516 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1517 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1518 [(set DstRC:$dst, (Int DstRC:$src1, SrcRC:$src2))]>;
1519 def rm : SI<opc, MRMSrcMem, (outs DstRC:$dst),
1520 (ins DstRC:$src1, x86memop:$src2),
1522 !strconcat(asm, "\t{$src2, $dst|$dst, $src2}"),
1523 !strconcat(asm, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
1524 [(set DstRC:$dst, (Int DstRC:$src1, (ld_frag addr:$src2)))]>;
1527 defm Int_VCVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1528 f128mem, load, "cvtsd2si">, XD, VEX;
1529 defm Int_VCVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64,
1530 int_x86_sse2_cvtsd2si64, f128mem, load, "cvtsd2si">,
1533 // FIXME: The asm matcher has a hack to ignore instructions with _Int and Int_
1534 // Get rid of this hack or rename the intrinsics, there are several
1535 // intructions that only match with the intrinsic form, why create duplicates
1536 // to let them be recognized by the assembler?
1537 defm VCVTSD2SI : sse12_cvt_s_np<0x2D, FR64, GR32, f64mem,
1538 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_LIG;
1539 defm VCVTSD2SI64 : sse12_cvt_s_np<0x2D, FR64, GR64, f64mem,
1540 "cvtsd2si\t{$src, $dst|$dst, $src}">, XD, VEX, VEX_W,
1543 defm CVTSD2SI : sse12_cvt_sint<0x2D, VR128, GR32, int_x86_sse2_cvtsd2si,
1544 f128mem, load, "cvtsd2si{l}">, XD;
1545 defm CVTSD2SI64 : sse12_cvt_sint<0x2D, VR128, GR64, int_x86_sse2_cvtsd2si64,
1546 f128mem, load, "cvtsd2si{q}">, XD, REX_W;
1549 defm Int_VCVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1550 int_x86_sse_cvtsi2ss, i32mem, loadi32, "cvtsi2ss", 0>, XS, VEX_4V;
1551 defm Int_VCVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1552 int_x86_sse_cvtsi642ss, i64mem, loadi64, "cvtsi2ss", 0>, XS, VEX_4V,
1554 defm Int_VCVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1555 int_x86_sse2_cvtsi2sd, i32mem, loadi32, "cvtsi2sd", 0>, XD, VEX_4V;
1556 defm Int_VCVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1557 int_x86_sse2_cvtsi642sd, i64mem, loadi64, "cvtsi2sd", 0>, XD,
1560 let Constraints = "$src1 = $dst" in {
1561 defm Int_CVTSI2SS : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1562 int_x86_sse_cvtsi2ss, i32mem, loadi32,
1564 defm Int_CVTSI2SS64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1565 int_x86_sse_cvtsi642ss, i64mem, loadi64,
1566 "cvtsi2ss{q}">, XS, REX_W;
1567 defm Int_CVTSI2SD : sse12_cvt_sint_3addr<0x2A, GR32, VR128,
1568 int_x86_sse2_cvtsi2sd, i32mem, loadi32,
1570 defm Int_CVTSI2SD64 : sse12_cvt_sint_3addr<0x2A, GR64, VR128,
1571 int_x86_sse2_cvtsi642sd, i64mem, loadi64,
1572 "cvtsi2sd">, XD, REX_W;
1577 // Aliases for intrinsics
1578 defm Int_VCVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1579 f32mem, load, "cvttss2si">, XS, VEX;
1580 defm Int_VCVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1581 int_x86_sse_cvttss2si64, f32mem, load,
1582 "cvttss2si">, XS, VEX, VEX_W;
1583 defm Int_VCVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1584 f128mem, load, "cvttsd2si">, XD, VEX;
1585 defm Int_VCVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1586 int_x86_sse2_cvttsd2si64, f128mem, load,
1587 "cvttsd2si">, XD, VEX, VEX_W;
1588 defm Int_CVTTSS2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse_cvttss2si,
1589 f32mem, load, "cvttss2si">, XS;
1590 defm Int_CVTTSS2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1591 int_x86_sse_cvttss2si64, f32mem, load,
1592 "cvttss2si{q}">, XS, REX_W;
1593 defm Int_CVTTSD2SI : sse12_cvt_sint<0x2C, VR128, GR32, int_x86_sse2_cvttsd2si,
1594 f128mem, load, "cvttsd2si">, XD;
1595 defm Int_CVTTSD2SI64 : sse12_cvt_sint<0x2C, VR128, GR64,
1596 int_x86_sse2_cvttsd2si64, f128mem, load,
1597 "cvttsd2si{q}">, XD, REX_W;
1599 let Pattern = []<dag> in {
1600 defm VCVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load,
1601 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS,
1603 defm VCVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load,
1604 "cvtss2si\t{$src, $dst|$dst, $src}">, XS, VEX,
1606 defm VCVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load,
1607 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1608 SSEPackedSingle>, TB, VEX;
1609 defm VCVTDQ2PSY : sse12_cvt_p<0x5B, VR256, VR256, undef, i256mem, load,
1610 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1611 SSEPackedSingle>, TB, VEX;
1614 let Pattern = []<dag> in {
1615 defm CVTSS2SI : sse12_cvt_s<0x2D, FR32, GR32, undef, f32mem, load /*dummy*/,
1616 "cvtss2si{l}\t{$src, $dst|$dst, $src}">, XS;
1617 defm CVTSS2SI64 : sse12_cvt_s<0x2D, FR32, GR64, undef, f32mem, load /*dummy*/,
1618 "cvtss2si{q}\t{$src, $dst|$dst, $src}">, XS, REX_W;
1619 defm CVTDQ2PS : sse12_cvt_p<0x5B, VR128, VR128, undef, i128mem, load /*dummy*/,
1620 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1621 SSEPackedSingle>, TB; /* PD SSE3 form is avaiable */
1624 let Predicates = [HasSSE1] in {
1625 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1626 (CVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1627 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1628 (CVTSS2SIrm addr:$src)>;
1629 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1630 (CVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1631 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1632 (CVTSS2SI64rm addr:$src)>;
1635 let Predicates = [HasAVX] in {
1636 def : Pat<(int_x86_sse_cvtss2si VR128:$src),
1637 (VCVTSS2SIrr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1638 def : Pat<(int_x86_sse_cvtss2si (load addr:$src)),
1639 (VCVTSS2SIrm addr:$src)>;
1640 def : Pat<(int_x86_sse_cvtss2si64 VR128:$src),
1641 (VCVTSS2SI64rr (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss))>;
1642 def : Pat<(int_x86_sse_cvtss2si64 (load addr:$src)),
1643 (VCVTSS2SI64rm addr:$src)>;
1648 // Convert scalar double to scalar single
1649 def VCVTSD2SSrr : VSDI<0x5A, MRMSrcReg, (outs FR32:$dst),
1650 (ins FR64:$src1, FR64:$src2),
1651 "cvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
1654 def VCVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst),
1655 (ins FR64:$src1, f64mem:$src2),
1656 "vcvtsd2ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1657 []>, XD, Requires<[HasAVX, OptForSize]>, VEX_4V, VEX_LIG;
1659 def : Pat<(f32 (fround FR64:$src)), (VCVTSD2SSrr FR64:$src, FR64:$src)>,
1662 def CVTSD2SSrr : SDI<0x5A, MRMSrcReg, (outs FR32:$dst), (ins FR64:$src),
1663 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1664 [(set FR32:$dst, (fround FR64:$src))]>;
1665 def CVTSD2SSrm : I<0x5A, MRMSrcMem, (outs FR32:$dst), (ins f64mem:$src),
1666 "cvtsd2ss\t{$src, $dst|$dst, $src}",
1667 [(set FR32:$dst, (fround (loadf64 addr:$src)))]>, XD,
1668 Requires<[HasSSE2, OptForSize]>;
1670 defm Int_VCVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1671 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss", 0>,
1673 let Constraints = "$src1 = $dst" in
1674 defm Int_CVTSD2SS: sse12_cvt_sint_3addr<0x5A, VR128, VR128,
1675 int_x86_sse2_cvtsd2ss, f64mem, load, "cvtsd2ss">, XS;
1677 // Convert scalar single to scalar double
1678 // SSE2 instructions with XS prefix
1679 def VCVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst),
1680 (ins FR32:$src1, FR32:$src2),
1681 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1682 []>, XS, Requires<[HasAVX]>, VEX_4V, VEX_LIG;
1684 def VCVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst),
1685 (ins FR32:$src1, f32mem:$src2),
1686 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1687 []>, XS, VEX_4V, VEX_LIG, Requires<[HasAVX, OptForSize]>;
1689 let Predicates = [HasAVX] in {
1690 def : Pat<(f64 (fextend FR32:$src)),
1691 (VCVTSS2SDrr FR32:$src, FR32:$src)>;
1692 def : Pat<(fextend (loadf32 addr:$src)),
1693 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1694 def : Pat<(extloadf32 addr:$src),
1695 (VCVTSS2SDrm (f32 (IMPLICIT_DEF)), addr:$src)>;
1698 def : Pat<(extloadf32 addr:$src),
1699 (VCVTSS2SDrr (f32 (IMPLICIT_DEF)), (MOVSSrm addr:$src))>,
1700 Requires<[HasAVX, OptForSpeed]>;
1702 def CVTSS2SDrr : I<0x5A, MRMSrcReg, (outs FR64:$dst), (ins FR32:$src),
1703 "cvtss2sd\t{$src, $dst|$dst, $src}",
1704 [(set FR64:$dst, (fextend FR32:$src))]>, XS,
1705 Requires<[HasSSE2]>;
1706 def CVTSS2SDrm : I<0x5A, MRMSrcMem, (outs FR64:$dst), (ins f32mem:$src),
1707 "cvtss2sd\t{$src, $dst|$dst, $src}",
1708 [(set FR64:$dst, (extloadf32 addr:$src))]>, XS,
1709 Requires<[HasSSE2, OptForSize]>;
1711 // extload f32 -> f64. This matches load+fextend because we have a hack in
1712 // the isel (PreprocessForFPConvert) that can introduce loads after dag
1714 // Since these loads aren't folded into the fextend, we have to match it
1716 def : Pat<(fextend (loadf32 addr:$src)),
1717 (CVTSS2SDrm addr:$src)>, Requires<[HasSSE2]>;
1718 def : Pat<(extloadf32 addr:$src),
1719 (CVTSS2SDrr (MOVSSrm addr:$src))>, Requires<[HasSSE2, OptForSpeed]>;
1721 def Int_VCVTSS2SDrr: I<0x5A, MRMSrcReg,
1722 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1723 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1724 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1725 VR128:$src2))]>, XS, VEX_4V,
1727 def Int_VCVTSS2SDrm: I<0x5A, MRMSrcMem,
1728 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1729 "vcvtss2sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
1730 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1731 (load addr:$src2)))]>, XS, VEX_4V,
1733 let Constraints = "$src1 = $dst" in { // SSE2 instructions with XS prefix
1734 def Int_CVTSS2SDrr: I<0x5A, MRMSrcReg,
1735 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
1736 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1737 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1738 VR128:$src2))]>, XS,
1739 Requires<[HasSSE2]>;
1740 def Int_CVTSS2SDrm: I<0x5A, MRMSrcMem,
1741 (outs VR128:$dst), (ins VR128:$src1, f32mem:$src2),
1742 "cvtss2sd\t{$src2, $dst|$dst, $src2}",
1743 [(set VR128:$dst, (int_x86_sse2_cvtss2sd VR128:$src1,
1744 (load addr:$src2)))]>, XS,
1745 Requires<[HasSSE2]>;
1748 // Convert doubleword to packed single/double fp
1749 // SSE2 instructions without OpSize prefix
1750 def Int_VCVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1751 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1752 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1753 TB, VEX, Requires<[HasAVX]>;
1754 def Int_VCVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1755 "vcvtdq2ps\t{$src, $dst|$dst, $src}",
1756 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1757 (bitconvert (memopv2i64 addr:$src))))]>,
1758 TB, VEX, Requires<[HasAVX]>;
1759 def Int_CVTDQ2PSrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1760 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1761 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps VR128:$src))]>,
1762 TB, Requires<[HasSSE2]>;
1763 def Int_CVTDQ2PSrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
1764 "cvtdq2ps\t{$src, $dst|$dst, $src}",
1765 [(set VR128:$dst, (int_x86_sse2_cvtdq2ps
1766 (bitconvert (memopv2i64 addr:$src))))]>,
1767 TB, Requires<[HasSSE2]>;
1769 // FIXME: why the non-intrinsic version is described as SSE3?
1770 // SSE2 instructions with XS prefix
1771 def Int_VCVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1772 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1773 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1774 XS, VEX, Requires<[HasAVX]>;
1775 def Int_VCVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1776 "vcvtdq2pd\t{$src, $dst|$dst, $src}",
1777 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1778 (bitconvert (memopv2i64 addr:$src))))]>,
1779 XS, VEX, Requires<[HasAVX]>;
1780 def Int_CVTDQ2PDrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1781 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1782 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd VR128:$src))]>,
1783 XS, Requires<[HasSSE2]>;
1784 def Int_CVTDQ2PDrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
1785 "cvtdq2pd\t{$src, $dst|$dst, $src}",
1786 [(set VR128:$dst, (int_x86_sse2_cvtdq2pd
1787 (bitconvert (memopv2i64 addr:$src))))]>,
1788 XS, Requires<[HasSSE2]>;
1791 // Convert packed single/double fp to doubleword
1792 def VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1793 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1794 def VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1795 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1796 def VCVTPS2DQYrr : VPDI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1797 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1798 def VCVTPS2DQYrm : VPDI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1799 "cvtps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1800 def CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1801 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1802 def CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1803 "cvtps2dq\t{$src, $dst|$dst, $src}", []>;
1805 def Int_VCVTPS2DQrr : VPDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1806 "cvtps2dq\t{$src, $dst|$dst, $src}",
1807 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>,
1809 def Int_VCVTPS2DQrm : VPDI<0x5B, MRMSrcMem, (outs VR128:$dst),
1811 "cvtps2dq\t{$src, $dst|$dst, $src}",
1812 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1813 (memop addr:$src)))]>, VEX;
1814 def Int_CVTPS2DQrr : PDI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1815 "cvtps2dq\t{$src, $dst|$dst, $src}",
1816 [(set VR128:$dst, (int_x86_sse2_cvtps2dq VR128:$src))]>;
1817 def Int_CVTPS2DQrm : PDI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1818 "cvtps2dq\t{$src, $dst|$dst, $src}",
1819 [(set VR128:$dst, (int_x86_sse2_cvtps2dq
1820 (memop addr:$src)))]>;
1822 // SSE2 packed instructions with XD prefix
1823 def Int_VCVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1824 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1825 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1826 XD, VEX, Requires<[HasAVX]>;
1827 def Int_VCVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1828 "vcvtpd2dq\t{$src, $dst|$dst, $src}",
1829 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1830 (memop addr:$src)))]>,
1831 XD, VEX, Requires<[HasAVX]>;
1832 def Int_CVTPD2DQrr : I<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1833 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1834 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq VR128:$src))]>,
1835 XD, Requires<[HasSSE2]>;
1836 def Int_CVTPD2DQrm : I<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1837 "cvtpd2dq\t{$src, $dst|$dst, $src}",
1838 [(set VR128:$dst, (int_x86_sse2_cvtpd2dq
1839 (memop addr:$src)))]>,
1840 XD, Requires<[HasSSE2]>;
1843 // Convert with truncation packed single/double fp to doubleword
1844 // SSE2 packed instructions with XS prefix
1845 def VCVTTPS2DQrr : VSSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1846 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1848 def VCVTTPS2DQrm : VSSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1849 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1850 def VCVTTPS2DQYrr : VSSI<0x5B, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
1851 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1853 def VCVTTPS2DQYrm : VSSI<0x5B, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
1854 "cvttps2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1855 def CVTTPS2DQrr : SSI<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1856 "cvttps2dq\t{$src, $dst|$dst, $src}",
1858 (int_x86_sse2_cvttps2dq VR128:$src))]>;
1859 def CVTTPS2DQrm : SSI<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1860 "cvttps2dq\t{$src, $dst|$dst, $src}",
1862 (int_x86_sse2_cvttps2dq (memop addr:$src)))]>;
1864 def Int_VCVTTPS2DQrr : I<0x5B, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1865 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1867 (int_x86_sse2_cvttps2dq VR128:$src))]>,
1868 XS, VEX, Requires<[HasAVX]>;
1869 def Int_VCVTTPS2DQrm : I<0x5B, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1870 "vcvttps2dq\t{$src, $dst|$dst, $src}",
1871 [(set VR128:$dst, (int_x86_sse2_cvttps2dq
1872 (memop addr:$src)))]>,
1873 XS, VEX, Requires<[HasAVX]>;
1875 let Predicates = [HasSSE2] in {
1876 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1877 (Int_CVTDQ2PSrr VR128:$src)>;
1878 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1879 (CVTTPS2DQrr VR128:$src)>;
1882 let Predicates = [HasAVX] in {
1883 def : Pat<(v4f32 (sint_to_fp (v4i32 VR128:$src))),
1884 (Int_VCVTDQ2PSrr VR128:$src)>;
1885 def : Pat<(v4i32 (fp_to_sint (v4f32 VR128:$src))),
1886 (VCVTTPS2DQrr VR128:$src)>;
1887 def : Pat<(v8f32 (sint_to_fp (v8i32 VR256:$src))),
1888 (VCVTDQ2PSYrr VR256:$src)>;
1889 def : Pat<(v8i32 (fp_to_sint (v8f32 VR256:$src))),
1890 (VCVTTPS2DQYrr VR256:$src)>;
1893 def VCVTTPD2DQrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1894 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1896 (int_x86_sse2_cvttpd2dq VR128:$src))]>, VEX;
1897 let isCodeGenOnly = 1 in
1898 def VCVTTPD2DQrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1899 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1900 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1901 (memop addr:$src)))]>, VEX;
1902 def CVTTPD2DQrr : PDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1903 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1904 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq VR128:$src))]>;
1905 def CVTTPD2DQrm : PDI<0xE6, MRMSrcMem, (outs VR128:$dst),(ins f128mem:$src),
1906 "cvttpd2dq\t{$src, $dst|$dst, $src}",
1907 [(set VR128:$dst, (int_x86_sse2_cvttpd2dq
1908 (memop addr:$src)))]>;
1910 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1911 // register, but the same isn't true when using memory operands instead.
1912 // Provide other assembly rr and rm forms to address this explicitly.
1913 def VCVTTPD2DQXrYr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1914 "cvttpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
1917 def VCVTTPD2DQXrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1918 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1919 def VCVTTPD2DQXrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1920 "cvttpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
1923 def VCVTTPD2DQYrr : VPDI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1924 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
1925 def VCVTTPD2DQYrm : VPDI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1926 "cvttpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1928 // Convert packed single to packed double
1929 let Predicates = [HasAVX] in {
1930 // SSE2 instructions without OpSize prefix
1931 def VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1932 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1933 def VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1934 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1935 def VCVTPS2PDYrr : I<0x5A, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
1936 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1937 def VCVTPS2PDYrm : I<0x5A, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
1938 "vcvtps2pd\t{$src, $dst|$dst, $src}", []>, TB, VEX;
1940 def CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1941 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1942 def CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1943 "cvtps2pd\t{$src, $dst|$dst, $src}", []>, TB;
1945 def Int_VCVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1946 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1947 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1948 TB, VEX, Requires<[HasAVX]>;
1949 def Int_VCVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1950 "vcvtps2pd\t{$src, $dst|$dst, $src}",
1951 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1952 (load addr:$src)))]>,
1953 TB, VEX, Requires<[HasAVX]>;
1954 def Int_CVTPS2PDrr : I<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1955 "cvtps2pd\t{$src, $dst|$dst, $src}",
1956 [(set VR128:$dst, (int_x86_sse2_cvtps2pd VR128:$src))]>,
1957 TB, Requires<[HasSSE2]>;
1958 def Int_CVTPS2PDrm : I<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
1959 "cvtps2pd\t{$src, $dst|$dst, $src}",
1960 [(set VR128:$dst, (int_x86_sse2_cvtps2pd
1961 (load addr:$src)))]>,
1962 TB, Requires<[HasSSE2]>;
1964 // Convert packed double to packed single
1965 // The assembler can recognize rr 256-bit instructions by seeing a ymm
1966 // register, but the same isn't true when using memory operands instead.
1967 // Provide other assembly rr and rm forms to address this explicitly.
1968 def VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1969 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1970 def VCVTPD2PSXrYr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1971 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>, VEX;
1974 def VCVTPD2PSXrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1975 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1976 def VCVTPD2PSXrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1977 "cvtpd2psx\t{$src, $dst|$dst, $src}", []>, VEX;
1980 def VCVTPD2PSYrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
1981 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX;
1982 def VCVTPD2PSYrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
1983 "cvtpd2psy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
1984 def CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1985 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1986 def CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
1987 "cvtpd2ps\t{$src, $dst|$dst, $src}", []>;
1990 def Int_VCVTPD2PSrr : VPDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1991 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1992 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
1993 def Int_VCVTPD2PSrm : VPDI<0x5A, MRMSrcMem, (outs VR128:$dst),
1995 "cvtpd2ps\t{$src, $dst|$dst, $src}",
1996 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
1997 (memop addr:$src)))]>;
1998 def Int_CVTPD2PSrr : PDI<0x5A, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
1999 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2000 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps VR128:$src))]>;
2001 def Int_CVTPD2PSrm : PDI<0x5A, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2002 "cvtpd2ps\t{$src, $dst|$dst, $src}",
2003 [(set VR128:$dst, (int_x86_sse2_cvtpd2ps
2004 (memop addr:$src)))]>;
2006 // AVX 256-bit register conversion intrinsics
2007 // FIXME: Migrate SSE conversion intrinsics matching to use patterns as below
2008 // whenever possible to avoid declaring two versions of each one.
2009 def : Pat<(int_x86_avx_cvtdq2_ps_256 VR256:$src),
2010 (VCVTDQ2PSYrr VR256:$src)>;
2011 def : Pat<(int_x86_avx_cvtdq2_ps_256 (bitconvert (memopv4i64 addr:$src))),
2012 (VCVTDQ2PSYrm addr:$src)>;
2014 def : Pat<(int_x86_avx_cvt_pd2_ps_256 VR256:$src),
2015 (VCVTPD2PSYrr VR256:$src)>;
2016 def : Pat<(int_x86_avx_cvt_pd2_ps_256 (memopv4f64 addr:$src)),
2017 (VCVTPD2PSYrm addr:$src)>;
2019 def : Pat<(int_x86_avx_cvt_ps2dq_256 VR256:$src),
2020 (VCVTPS2DQYrr VR256:$src)>;
2021 def : Pat<(int_x86_avx_cvt_ps2dq_256 (memopv8f32 addr:$src)),
2022 (VCVTPS2DQYrm addr:$src)>;
2024 def : Pat<(int_x86_avx_cvt_ps2_pd_256 VR128:$src),
2025 (VCVTPS2PDYrr VR128:$src)>;
2026 def : Pat<(int_x86_avx_cvt_ps2_pd_256 (memopv4f32 addr:$src)),
2027 (VCVTPS2PDYrm addr:$src)>;
2029 def : Pat<(int_x86_avx_cvtt_pd2dq_256 VR256:$src),
2030 (VCVTTPD2DQYrr VR256:$src)>;
2031 def : Pat<(int_x86_avx_cvtt_pd2dq_256 (memopv4f64 addr:$src)),
2032 (VCVTTPD2DQYrm addr:$src)>;
2034 def : Pat<(int_x86_avx_cvtt_ps2dq_256 VR256:$src),
2035 (VCVTTPS2DQYrr VR256:$src)>;
2036 def : Pat<(int_x86_avx_cvtt_ps2dq_256 (memopv8f32 addr:$src)),
2037 (VCVTTPS2DQYrm addr:$src)>;
2039 // Match fround and fextend for 128/256-bit conversions
2040 def : Pat<(v4f32 (fround (v4f64 VR256:$src))),
2041 (VCVTPD2PSYrr VR256:$src)>;
2042 def : Pat<(v4f32 (fround (loadv4f64 addr:$src))),
2043 (VCVTPD2PSYrm addr:$src)>;
2045 def : Pat<(v4f64 (fextend (v4f32 VR128:$src))),
2046 (VCVTPS2PDYrr VR128:$src)>;
2047 def : Pat<(v4f64 (fextend (loadv4f32 addr:$src))),
2048 (VCVTPS2PDYrm addr:$src)>;
2050 //===----------------------------------------------------------------------===//
2051 // SSE 1 & 2 - Compare Instructions
2052 //===----------------------------------------------------------------------===//
2054 // sse12_cmp_scalar - sse 1 & 2 compare scalar instructions
2055 multiclass sse12_cmp_scalar<RegisterClass RC, X86MemOperand x86memop,
2056 SDNode OpNode, ValueType VT, PatFrag ld_frag,
2057 string asm, string asm_alt> {
2058 def rr : SIi8<0xC2, MRMSrcReg,
2059 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2060 [(set RC:$dst, (OpNode (VT RC:$src1), RC:$src2, imm:$cc))]>;
2061 def rm : SIi8<0xC2, MRMSrcMem,
2062 (outs RC:$dst), (ins RC:$src1, x86memop:$src2, SSECC:$cc), asm,
2063 [(set RC:$dst, (OpNode (VT RC:$src1),
2064 (ld_frag addr:$src2), imm:$cc))]>;
2066 // Accept explicit immediate argument form instead of comparison code.
2067 let neverHasSideEffects = 1 in {
2068 def rr_alt : SIi8<0xC2, MRMSrcReg, (outs RC:$dst),
2069 (ins RC:$src1, RC:$src2, i8imm:$cc), asm_alt, []>;
2071 def rm_alt : SIi8<0xC2, MRMSrcMem, (outs RC:$dst),
2072 (ins RC:$src1, x86memop:$src2, i8imm:$cc), asm_alt, []>;
2076 defm VCMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2077 "cmp${cc}ss\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2078 "cmpss\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2079 XS, VEX_4V, VEX_LIG;
2080 defm VCMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2081 "cmp${cc}sd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2082 "cmpsd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}">,
2083 XD, VEX_4V, VEX_LIG;
2085 let Constraints = "$src1 = $dst" in {
2086 defm CMPSS : sse12_cmp_scalar<FR32, f32mem, X86cmpss, f32, loadf32,
2087 "cmp${cc}ss\t{$src2, $dst|$dst, $src2}",
2088 "cmpss\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2090 defm CMPSD : sse12_cmp_scalar<FR64, f64mem, X86cmpsd, f64, loadf64,
2091 "cmp${cc}sd\t{$src2, $dst|$dst, $src2}",
2092 "cmpsd\t{$cc, $src2, $dst|$dst, $src2, $cc}">,
2096 multiclass sse12_cmp_scalar_int<RegisterClass RC, X86MemOperand x86memop,
2097 Intrinsic Int, string asm> {
2098 def rr : SIi8<0xC2, MRMSrcReg, (outs VR128:$dst),
2099 (ins VR128:$src1, VR128:$src, SSECC:$cc), asm,
2100 [(set VR128:$dst, (Int VR128:$src1,
2101 VR128:$src, imm:$cc))]>;
2102 def rm : SIi8<0xC2, MRMSrcMem, (outs VR128:$dst),
2103 (ins VR128:$src1, f32mem:$src, SSECC:$cc), asm,
2104 [(set VR128:$dst, (Int VR128:$src1,
2105 (load addr:$src), imm:$cc))]>;
2108 // Aliases to match intrinsics which expect XMM operand(s).
2109 defm Int_VCMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2110 "cmp${cc}ss\t{$src, $src1, $dst|$dst, $src1, $src}">,
2112 defm Int_VCMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2113 "cmp${cc}sd\t{$src, $src1, $dst|$dst, $src1, $src}">,
2115 let Constraints = "$src1 = $dst" in {
2116 defm Int_CMPSS : sse12_cmp_scalar_int<VR128, f32mem, int_x86_sse_cmp_ss,
2117 "cmp${cc}ss\t{$src, $dst|$dst, $src}">, XS;
2118 defm Int_CMPSD : sse12_cmp_scalar_int<VR128, f64mem, int_x86_sse2_cmp_sd,
2119 "cmp${cc}sd\t{$src, $dst|$dst, $src}">, XD;
2123 // sse12_ord_cmp - Unordered/Ordered scalar fp compare and set EFLAGS
2124 multiclass sse12_ord_cmp<bits<8> opc, RegisterClass RC, SDNode OpNode,
2125 ValueType vt, X86MemOperand x86memop,
2126 PatFrag ld_frag, string OpcodeStr, Domain d> {
2127 def rr: PI<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
2128 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2129 [(set EFLAGS, (OpNode (vt RC:$src1), RC:$src2))], d>;
2130 def rm: PI<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
2131 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
2132 [(set EFLAGS, (OpNode (vt RC:$src1),
2133 (ld_frag addr:$src2)))], d>;
2136 let Defs = [EFLAGS] in {
2137 defm VUCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2138 "ucomiss", SSEPackedSingle>, TB, VEX, VEX_LIG;
2139 defm VUCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2140 "ucomisd", SSEPackedDouble>, TB, OpSize, VEX,
2142 let Pattern = []<dag> in {
2143 defm VCOMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2144 "comiss", SSEPackedSingle>, TB, VEX,
2146 defm VCOMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2147 "comisd", SSEPackedDouble>, TB, OpSize, VEX,
2151 defm Int_VUCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2152 load, "ucomiss", SSEPackedSingle>, TB, VEX;
2153 defm Int_VUCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2154 load, "ucomisd", SSEPackedDouble>, TB, OpSize, VEX;
2156 defm Int_VCOMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem,
2157 load, "comiss", SSEPackedSingle>, TB, VEX;
2158 defm Int_VCOMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem,
2159 load, "comisd", SSEPackedDouble>, TB, OpSize, VEX;
2160 defm UCOMISS : sse12_ord_cmp<0x2E, FR32, X86cmp, f32, f32mem, loadf32,
2161 "ucomiss", SSEPackedSingle>, TB;
2162 defm UCOMISD : sse12_ord_cmp<0x2E, FR64, X86cmp, f64, f64mem, loadf64,
2163 "ucomisd", SSEPackedDouble>, TB, OpSize;
2165 let Pattern = []<dag> in {
2166 defm COMISS : sse12_ord_cmp<0x2F, VR128, undef, v4f32, f128mem, load,
2167 "comiss", SSEPackedSingle>, TB;
2168 defm COMISD : sse12_ord_cmp<0x2F, VR128, undef, v2f64, f128mem, load,
2169 "comisd", SSEPackedDouble>, TB, OpSize;
2172 defm Int_UCOMISS : sse12_ord_cmp<0x2E, VR128, X86ucomi, v4f32, f128mem,
2173 load, "ucomiss", SSEPackedSingle>, TB;
2174 defm Int_UCOMISD : sse12_ord_cmp<0x2E, VR128, X86ucomi, v2f64, f128mem,
2175 load, "ucomisd", SSEPackedDouble>, TB, OpSize;
2177 defm Int_COMISS : sse12_ord_cmp<0x2F, VR128, X86comi, v4f32, f128mem, load,
2178 "comiss", SSEPackedSingle>, TB;
2179 defm Int_COMISD : sse12_ord_cmp<0x2F, VR128, X86comi, v2f64, f128mem, load,
2180 "comisd", SSEPackedDouble>, TB, OpSize;
2181 } // Defs = [EFLAGS]
2183 // sse12_cmp_packed - sse 1 & 2 compared packed instructions
2184 multiclass sse12_cmp_packed<RegisterClass RC, X86MemOperand x86memop,
2185 Intrinsic Int, string asm, string asm_alt,
2187 let isAsmParserOnly = 1 in {
2188 def rri : PIi8<0xC2, MRMSrcReg,
2189 (outs RC:$dst), (ins RC:$src1, RC:$src2, SSECC:$cc), asm,
2190 [(set RC:$dst, (Int RC:$src1, RC:$src2, imm:$cc))], d>;
2191 def rmi : PIi8<0xC2, MRMSrcMem,
2192 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, SSECC:$cc), asm,
2193 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2), imm:$cc))], d>;
2196 // Accept explicit immediate argument form instead of comparison code.
2197 def rri_alt : PIi8<0xC2, MRMSrcReg,
2198 (outs RC:$dst), (ins RC:$src1, RC:$src2, i8imm:$cc),
2200 def rmi_alt : PIi8<0xC2, MRMSrcMem,
2201 (outs RC:$dst), (ins RC:$src1, f128mem:$src2, i8imm:$cc),
2205 defm VCMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2206 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2207 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2208 SSEPackedSingle>, TB, VEX_4V;
2209 defm VCMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2210 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2211 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2212 SSEPackedDouble>, TB, OpSize, VEX_4V;
2213 defm VCMPPSY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_ps_256,
2214 "cmp${cc}ps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2215 "cmpps\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2216 SSEPackedSingle>, TB, VEX_4V;
2217 defm VCMPPDY : sse12_cmp_packed<VR256, f256mem, int_x86_avx_cmp_pd_256,
2218 "cmp${cc}pd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2219 "cmppd\t{$cc, $src2, $src1, $dst|$dst, $src1, $src2, $cc}",
2220 SSEPackedDouble>, TB, OpSize, VEX_4V;
2221 let Constraints = "$src1 = $dst" in {
2222 defm CMPPS : sse12_cmp_packed<VR128, f128mem, int_x86_sse_cmp_ps,
2223 "cmp${cc}ps\t{$src2, $dst|$dst, $src2}",
2224 "cmpps\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2225 SSEPackedSingle>, TB;
2226 defm CMPPD : sse12_cmp_packed<VR128, f128mem, int_x86_sse2_cmp_pd,
2227 "cmp${cc}pd\t{$src2, $dst|$dst, $src2}",
2228 "cmppd\t{$cc, $src2, $dst|$dst, $src2, $cc}",
2229 SSEPackedDouble>, TB, OpSize;
2232 let Predicates = [HasSSE1] in {
2233 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2234 (CMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2235 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2236 (CMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2239 let Predicates = [HasSSE2] in {
2240 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2241 (CMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2242 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2243 (CMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2246 let Predicates = [HasAVX] in {
2247 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), VR128:$src2, imm:$cc)),
2248 (VCMPPSrri (v4f32 VR128:$src1), (v4f32 VR128:$src2), imm:$cc)>;
2249 def : Pat<(v4i32 (X86cmpps (v4f32 VR128:$src1), (memop addr:$src2), imm:$cc)),
2250 (VCMPPSrmi (v4f32 VR128:$src1), addr:$src2, imm:$cc)>;
2251 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), VR128:$src2, imm:$cc)),
2252 (VCMPPDrri VR128:$src1, VR128:$src2, imm:$cc)>;
2253 def : Pat<(v2i64 (X86cmppd (v2f64 VR128:$src1), (memop addr:$src2), imm:$cc)),
2254 (VCMPPDrmi VR128:$src1, addr:$src2, imm:$cc)>;
2256 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), VR256:$src2, imm:$cc)),
2257 (VCMPPSYrri (v8f32 VR256:$src1), (v8f32 VR256:$src2), imm:$cc)>;
2258 def : Pat<(v8i32 (X86cmpps (v8f32 VR256:$src1), (memop addr:$src2), imm:$cc)),
2259 (VCMPPSYrmi (v8f32 VR256:$src1), addr:$src2, imm:$cc)>;
2260 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), VR256:$src2, imm:$cc)),
2261 (VCMPPDYrri VR256:$src1, VR256:$src2, imm:$cc)>;
2262 def : Pat<(v4i64 (X86cmppd (v4f64 VR256:$src1), (memop addr:$src2), imm:$cc)),
2263 (VCMPPDYrmi VR256:$src1, addr:$src2, imm:$cc)>;
2266 //===----------------------------------------------------------------------===//
2267 // SSE 1 & 2 - Shuffle Instructions
2268 //===----------------------------------------------------------------------===//
2270 /// sse12_shuffle - sse 1 & 2 shuffle instructions
2271 multiclass sse12_shuffle<RegisterClass RC, X86MemOperand x86memop,
2272 ValueType vt, string asm, PatFrag mem_frag,
2273 Domain d, bit IsConvertibleToThreeAddress = 0> {
2274 def rmi : PIi8<0xC6, MRMSrcMem, (outs RC:$dst),
2275 (ins RC:$src1, f128mem:$src2, i8imm:$src3), asm,
2276 [(set RC:$dst, (vt (shufp:$src3
2277 RC:$src1, (mem_frag addr:$src2))))], d>;
2278 let isConvertibleToThreeAddress = IsConvertibleToThreeAddress in
2279 def rri : PIi8<0xC6, MRMSrcReg, (outs RC:$dst),
2280 (ins RC:$src1, RC:$src2, i8imm:$src3), asm,
2282 (vt (shufp:$src3 RC:$src1, RC:$src2)))], d>;
2285 defm VSHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2286 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2287 memopv4f32, SSEPackedSingle>, TB, VEX_4V;
2288 defm VSHUFPSY : sse12_shuffle<VR256, f256mem, v8f32,
2289 "shufps\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
2290 memopv8f32, SSEPackedSingle>, TB, VEX_4V;
2291 defm VSHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2292 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2293 memopv2f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2294 defm VSHUFPDY : sse12_shuffle<VR256, f256mem, v4f64,
2295 "shufpd\t{$src3, $src2, $src1, $dst|$dst, $src2, $src2, $src3}",
2296 memopv4f64, SSEPackedDouble>, TB, OpSize, VEX_4V;
2298 let Constraints = "$src1 = $dst" in {
2299 defm SHUFPS : sse12_shuffle<VR128, f128mem, v4f32,
2300 "shufps\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2301 memopv4f32, SSEPackedSingle, 1 /* cvt to pshufd */>,
2303 defm SHUFPD : sse12_shuffle<VR128, f128mem, v2f64,
2304 "shufpd\t{$src3, $src2, $dst|$dst, $src2, $src3}",
2305 memopv2f64, SSEPackedDouble>, TB, OpSize;
2308 let Predicates = [HasSSE1] in {
2309 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2310 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2311 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2312 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2313 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2314 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2315 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2316 (SHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2317 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2318 (SHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2319 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2320 // fall back to this for SSE1)
2321 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2322 (SHUFPSrri VR128:$src2, VR128:$src1,
2323 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2324 // Special unary SHUFPSrri case.
2325 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2326 (SHUFPSrri VR128:$src1, VR128:$src1,
2327 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2330 let Predicates = [HasSSE2] in {
2331 // Special binary v4i32 shuffle cases with SHUFPS.
2332 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2333 (SHUFPSrri VR128:$src1, VR128:$src2,
2334 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2335 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2336 (bc_v4i32 (memopv2i64 addr:$src2)))),
2337 (SHUFPSrmi VR128:$src1, addr:$src2,
2338 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2339 // Special unary SHUFPDrri cases.
2340 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2341 (SHUFPDrri VR128:$src1, VR128:$src1,
2342 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2343 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2344 (SHUFPDrri VR128:$src1, VR128:$src1,
2345 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2346 // Special binary v2i64 shuffle cases using SHUFPDrri.
2347 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2348 (SHUFPDrri VR128:$src1, VR128:$src2,
2349 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2350 // Generic SHUFPD patterns
2351 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2352 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2353 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2354 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2355 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2356 (SHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2357 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2358 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2359 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2360 (SHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2363 let Predicates = [HasAVX] in {
2364 def : Pat<(v4f32 (X86Shufp VR128:$src1,
2365 (memopv4f32 addr:$src2), (i8 imm:$imm))),
2366 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2367 def : Pat<(v4f32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2368 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2369 def : Pat<(v4i32 (X86Shufp VR128:$src1,
2370 (bc_v4i32 (memopv2i64 addr:$src2)), (i8 imm:$imm))),
2371 (VSHUFPSrmi VR128:$src1, addr:$src2, imm:$imm)>;
2372 def : Pat<(v4i32 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2373 (VSHUFPSrri VR128:$src1, VR128:$src2, imm:$imm)>;
2374 // vector_shuffle v1, v2 <4, 5, 2, 3> using SHUFPSrri (we prefer movsd, but
2375 // fall back to this for SSE1)
2376 def : Pat<(v4f32 (movlp:$src3 VR128:$src1, (v4f32 VR128:$src2))),
2377 (VSHUFPSrri VR128:$src2, VR128:$src1,
2378 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2379 // Special unary SHUFPSrri case.
2380 def : Pat<(v4f32 (pshufd:$src3 VR128:$src1, (undef))),
2381 (VSHUFPSrri VR128:$src1, VR128:$src1,
2382 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2383 // Special binary v4i32 shuffle cases with SHUFPS.
2384 def : Pat<(v4i32 (shufp:$src3 VR128:$src1, (v4i32 VR128:$src2))),
2385 (VSHUFPSrri VR128:$src1, VR128:$src2,
2386 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2387 def : Pat<(v4i32 (shufp:$src3 VR128:$src1,
2388 (bc_v4i32 (memopv2i64 addr:$src2)))),
2389 (VSHUFPSrmi VR128:$src1, addr:$src2,
2390 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2391 // Special unary SHUFPDrri cases.
2392 def : Pat<(v2i64 (pshufd:$src3 VR128:$src1, (undef))),
2393 (VSHUFPDrri VR128:$src1, VR128:$src1,
2394 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2395 def : Pat<(v2f64 (pshufd:$src3 VR128:$src1, (undef))),
2396 (VSHUFPDrri VR128:$src1, VR128:$src1,
2397 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2398 // Special binary v2i64 shuffle cases using SHUFPDrri.
2399 def : Pat<(v2i64 (shufp:$src3 VR128:$src1, VR128:$src2)),
2400 (VSHUFPDrri VR128:$src1, VR128:$src2,
2401 (SHUFFLE_get_shuf_imm VR128:$src3))>;
2403 def : Pat<(v2i64 (X86Shufp VR128:$src1,
2404 (memopv2i64 addr:$src2), (i8 imm:$imm))),
2405 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2406 def : Pat<(v2f64 (X86Shufp VR128:$src1,
2407 (memopv2f64 addr:$src2), (i8 imm:$imm))),
2408 (VSHUFPDrmi VR128:$src1, addr:$src2, imm:$imm)>;
2409 def : Pat<(v2i64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2410 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2411 def : Pat<(v2f64 (X86Shufp VR128:$src1, VR128:$src2, (i8 imm:$imm))),
2412 (VSHUFPDrri VR128:$src1, VR128:$src2, imm:$imm)>;
2415 def : Pat<(v8i32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2416 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2417 def : Pat<(v8i32 (X86Shufp VR256:$src1,
2418 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
2419 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2421 def : Pat<(v8f32 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2422 (VSHUFPSYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2423 def : Pat<(v8f32 (X86Shufp VR256:$src1,
2424 (memopv8f32 addr:$src2), (i8 imm:$imm))),
2425 (VSHUFPSYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2427 def : Pat<(v4i64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2428 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2429 def : Pat<(v4i64 (X86Shufp VR256:$src1,
2430 (memopv4i64 addr:$src2), (i8 imm:$imm))),
2431 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2433 def : Pat<(v4f64 (X86Shufp VR256:$src1, VR256:$src2, (i8 imm:$imm))),
2434 (VSHUFPDYrri VR256:$src1, VR256:$src2, imm:$imm)>;
2435 def : Pat<(v4f64 (X86Shufp VR256:$src1,
2436 (memopv4f64 addr:$src2), (i8 imm:$imm))),
2437 (VSHUFPDYrmi VR256:$src1, addr:$src2, imm:$imm)>;
2440 //===----------------------------------------------------------------------===//
2441 // SSE 1 & 2 - Unpack Instructions
2442 //===----------------------------------------------------------------------===//
2444 /// sse12_unpack_interleave - sse 1 & 2 unpack and interleave
2445 multiclass sse12_unpack_interleave<bits<8> opc, PatFrag OpNode, ValueType vt,
2446 PatFrag mem_frag, RegisterClass RC,
2447 X86MemOperand x86memop, string asm,
2449 def rr : PI<opc, MRMSrcReg,
2450 (outs RC:$dst), (ins RC:$src1, RC:$src2),
2452 (vt (OpNode RC:$src1, RC:$src2)))], d>;
2453 def rm : PI<opc, MRMSrcMem,
2454 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
2456 (vt (OpNode RC:$src1,
2457 (mem_frag addr:$src2))))], d>;
2460 let AddedComplexity = 10 in {
2461 defm VUNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2462 VR128, f128mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2463 SSEPackedSingle>, TB, VEX_4V;
2464 defm VUNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2465 VR128, f128mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2466 SSEPackedDouble>, TB, OpSize, VEX_4V;
2467 defm VUNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2468 VR128, f128mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2469 SSEPackedSingle>, TB, VEX_4V;
2470 defm VUNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2471 VR128, f128mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2472 SSEPackedDouble>, TB, OpSize, VEX_4V;
2474 defm VUNPCKHPSY: sse12_unpack_interleave<0x15, unpckh, v8f32, memopv8f32,
2475 VR256, f256mem, "unpckhps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2476 SSEPackedSingle>, TB, VEX_4V;
2477 defm VUNPCKHPDY: sse12_unpack_interleave<0x15, unpckh, v4f64, memopv4f64,
2478 VR256, f256mem, "unpckhpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2479 SSEPackedDouble>, TB, OpSize, VEX_4V;
2480 defm VUNPCKLPSY: sse12_unpack_interleave<0x14, unpckl, v8f32, memopv8f32,
2481 VR256, f256mem, "unpcklps\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2482 SSEPackedSingle>, TB, VEX_4V;
2483 defm VUNPCKLPDY: sse12_unpack_interleave<0x14, unpckl, v4f64, memopv4f64,
2484 VR256, f256mem, "unpcklpd\t{$src2, $src1, $dst|$dst, $src1, $src2}",
2485 SSEPackedDouble>, TB, OpSize, VEX_4V;
2487 let Constraints = "$src1 = $dst" in {
2488 defm UNPCKHPS: sse12_unpack_interleave<0x15, unpckh, v4f32, memopv4f32,
2489 VR128, f128mem, "unpckhps\t{$src2, $dst|$dst, $src2}",
2490 SSEPackedSingle>, TB;
2491 defm UNPCKHPD: sse12_unpack_interleave<0x15, unpckh, v2f64, memopv2f64,
2492 VR128, f128mem, "unpckhpd\t{$src2, $dst|$dst, $src2}",
2493 SSEPackedDouble>, TB, OpSize;
2494 defm UNPCKLPS: sse12_unpack_interleave<0x14, unpckl, v4f32, memopv4f32,
2495 VR128, f128mem, "unpcklps\t{$src2, $dst|$dst, $src2}",
2496 SSEPackedSingle>, TB;
2497 defm UNPCKLPD: sse12_unpack_interleave<0x14, unpckl, v2f64, memopv2f64,
2498 VR128, f128mem, "unpcklpd\t{$src2, $dst|$dst, $src2}",
2499 SSEPackedDouble>, TB, OpSize;
2500 } // Constraints = "$src1 = $dst"
2501 } // AddedComplexity
2503 let Predicates = [HasSSE1] in {
2504 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2505 (UNPCKLPSrm VR128:$src1, addr:$src2)>;
2506 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2507 (UNPCKLPSrr VR128:$src1, VR128:$src2)>;
2508 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2509 (UNPCKHPSrm VR128:$src1, addr:$src2)>;
2510 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2511 (UNPCKHPSrr VR128:$src1, VR128:$src2)>;
2514 let Predicates = [HasSSE2] in {
2515 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2516 (UNPCKLPDrm VR128:$src1, addr:$src2)>;
2517 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2518 (UNPCKLPDrr VR128:$src1, VR128:$src2)>;
2519 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2520 (UNPCKHPDrm VR128:$src1, addr:$src2)>;
2521 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2522 (UNPCKHPDrr VR128:$src1, VR128:$src2)>;
2524 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2525 // problem is during lowering, where it's not possible to recognize the load
2526 // fold cause it has two uses through a bitcast. One use disappears at isel
2527 // time and the fold opportunity reappears.
2528 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2529 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2531 let AddedComplexity = 10 in
2532 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2533 (UNPCKLPDrr VR128:$src, VR128:$src)>;
2536 let Predicates = [HasAVX] in {
2537 def : Pat<(v4f32 (X86Unpckl VR128:$src1, (memopv4f32 addr:$src2))),
2538 (VUNPCKLPSrm VR128:$src1, addr:$src2)>;
2539 def : Pat<(v4f32 (X86Unpckl VR128:$src1, VR128:$src2)),
2540 (VUNPCKLPSrr VR128:$src1, VR128:$src2)>;
2541 def : Pat<(v4f32 (X86Unpckh VR128:$src1, (memopv4f32 addr:$src2))),
2542 (VUNPCKHPSrm VR128:$src1, addr:$src2)>;
2543 def : Pat<(v4f32 (X86Unpckh VR128:$src1, VR128:$src2)),
2544 (VUNPCKHPSrr VR128:$src1, VR128:$src2)>;
2546 def : Pat<(v8f32 (X86Unpckl VR256:$src1, (memopv8f32 addr:$src2))),
2547 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
2548 def : Pat<(v8f32 (X86Unpckl VR256:$src1, VR256:$src2)),
2549 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
2550 def : Pat<(v8f32 (X86Unpckh VR256:$src1, (memopv8f32 addr:$src2))),
2551 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
2552 def : Pat<(v8f32 (X86Unpckh VR256:$src1, VR256:$src2)),
2553 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
2555 def : Pat<(v2f64 (X86Unpckl VR128:$src1, (memopv2f64 addr:$src2))),
2556 (VUNPCKLPDrm VR128:$src1, addr:$src2)>;
2557 def : Pat<(v2f64 (X86Unpckl VR128:$src1, VR128:$src2)),
2558 (VUNPCKLPDrr VR128:$src1, VR128:$src2)>;
2559 def : Pat<(v2f64 (X86Unpckh VR128:$src1, (memopv2f64 addr:$src2))),
2560 (VUNPCKHPDrm VR128:$src1, addr:$src2)>;
2561 def : Pat<(v2f64 (X86Unpckh VR128:$src1, VR128:$src2)),
2562 (VUNPCKHPDrr VR128:$src1, VR128:$src2)>;
2564 def : Pat<(v4f64 (X86Unpckl VR256:$src1, (memopv4f64 addr:$src2))),
2565 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
2566 def : Pat<(v4f64 (X86Unpckl VR256:$src1, VR256:$src2)),
2567 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
2568 def : Pat<(v4f64 (X86Unpckh VR256:$src1, (memopv4f64 addr:$src2))),
2569 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
2570 def : Pat<(v4f64 (X86Unpckh VR256:$src1, VR256:$src2)),
2571 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
2573 // FIXME: Instead of X86Movddup, there should be a X86Unpckl here, the
2574 // problem is during lowering, where it's not possible to recognize the load
2575 // fold cause it has two uses through a bitcast. One use disappears at isel
2576 // time and the fold opportunity reappears.
2577 def : Pat<(v2f64 (X86Movddup VR128:$src)),
2578 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2579 let AddedComplexity = 10 in
2580 def : Pat<(splat_lo (v2f64 VR128:$src), (undef)),
2581 (VUNPCKLPDrr VR128:$src, VR128:$src)>;
2584 //===----------------------------------------------------------------------===//
2585 // SSE 1 & 2 - Extract Floating-Point Sign mask
2586 //===----------------------------------------------------------------------===//
2588 /// sse12_extr_sign_mask - sse 1 & 2 unpack and interleave
2589 multiclass sse12_extr_sign_mask<RegisterClass RC, Intrinsic Int, string asm,
2591 def rr32 : PI<0x50, MRMSrcReg, (outs GR32:$dst), (ins RC:$src),
2592 !strconcat(asm, "\t{$src, $dst|$dst, $src}"),
2593 [(set GR32:$dst, (Int RC:$src))], d>;
2594 def rr64 : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins RC:$src),
2595 !strconcat(asm, "\t{$src, $dst|$dst, $src}"), [], d>, REX_W;
2598 defm MOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps, "movmskps",
2599 SSEPackedSingle>, TB;
2600 defm MOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd, "movmskpd",
2601 SSEPackedDouble>, TB, OpSize;
2603 def : Pat<(i32 (X86fgetsign FR32:$src)),
2604 (MOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2605 sub_ss))>, Requires<[HasSSE1]>;
2606 def : Pat<(i64 (X86fgetsign FR32:$src)),
2607 (MOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2608 sub_ss))>, Requires<[HasSSE1]>;
2609 def : Pat<(i32 (X86fgetsign FR64:$src)),
2610 (MOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2611 sub_sd))>, Requires<[HasSSE2]>;
2612 def : Pat<(i64 (X86fgetsign FR64:$src)),
2613 (MOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2614 sub_sd))>, Requires<[HasSSE2]>;
2616 let Predicates = [HasAVX] in {
2617 defm VMOVMSKPS : sse12_extr_sign_mask<VR128, int_x86_sse_movmsk_ps,
2618 "movmskps", SSEPackedSingle>, TB, VEX;
2619 defm VMOVMSKPD : sse12_extr_sign_mask<VR128, int_x86_sse2_movmsk_pd,
2620 "movmskpd", SSEPackedDouble>, TB,
2622 defm VMOVMSKPSY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_ps_256,
2623 "movmskps", SSEPackedSingle>, TB, VEX;
2624 defm VMOVMSKPDY : sse12_extr_sign_mask<VR256, int_x86_avx_movmsk_pd_256,
2625 "movmskpd", SSEPackedDouble>, TB,
2628 def : Pat<(i32 (X86fgetsign FR32:$src)),
2629 (VMOVMSKPSrr32 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2631 def : Pat<(i64 (X86fgetsign FR32:$src)),
2632 (VMOVMSKPSrr64 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)), FR32:$src,
2634 def : Pat<(i32 (X86fgetsign FR64:$src)),
2635 (VMOVMSKPDrr32 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2637 def : Pat<(i64 (X86fgetsign FR64:$src)),
2638 (VMOVMSKPDrr64 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)), FR64:$src,
2642 def VMOVMSKPSr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2643 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2644 def VMOVMSKPDr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
2645 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2647 def VMOVMSKPSYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2648 "movmskps\t{$src, $dst|$dst, $src}", [], SSEPackedSingle>, TB, VEX;
2649 def VMOVMSKPDYr64r : PI<0x50, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
2650 "movmskpd\t{$src, $dst|$dst, $src}", [], SSEPackedDouble>, TB,
2654 //===----------------------------------------------------------------------===//
2655 // SSE 1 & 2 - Logical Instructions
2656 //===----------------------------------------------------------------------===//
2658 /// sse12_fp_alias_pack_logical - SSE 1 & 2 aliased packed FP logical ops
2660 multiclass sse12_fp_alias_pack_logical<bits<8> opc, string OpcodeStr,
2662 defm V#NAME#PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode,
2663 FR32, f32, f128mem, memopfsf32, SSEPackedSingle, 0>, TB, VEX_4V;
2665 defm V#NAME#PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode,
2666 FR64, f64, f128mem, memopfsf64, SSEPackedDouble, 0>, TB, OpSize, VEX_4V;
2668 let Constraints = "$src1 = $dst" in {
2669 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, FR32,
2670 f32, f128mem, memopfsf32, SSEPackedSingle>, TB;
2672 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, FR64,
2673 f64, f128mem, memopfsf64, SSEPackedDouble>, TB, OpSize;
2677 // Alias bitwise logical operations using SSE logical ops on packed FP values.
2678 let mayLoad = 0 in {
2679 defm FsAND : sse12_fp_alias_pack_logical<0x54, "and", X86fand>;
2680 defm FsOR : sse12_fp_alias_pack_logical<0x56, "or", X86for>;
2681 defm FsXOR : sse12_fp_alias_pack_logical<0x57, "xor", X86fxor>;
2684 let neverHasSideEffects = 1, Pattern = []<dag>, isCommutable = 0 in
2685 defm FsANDN : sse12_fp_alias_pack_logical<0x55, "andn", undef>;
2687 /// sse12_fp_packed_logical - SSE 1 & 2 packed FP logical ops
2689 multiclass sse12_fp_packed_logical<bits<8> opc, string OpcodeStr,
2691 // In AVX no need to add a pattern for 128-bit logical rr ps, because they
2692 // are all promoted to v2i64, and the patterns are covered by the int
2693 // version. This is needed in SSE only, because v2i64 isn't supported on
2694 // SSE1, but only on SSE2.
2695 defm V#NAME#PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2696 !strconcat(OpcodeStr, "ps"), f128mem, [],
2697 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2698 (memopv2i64 addr:$src2)))], 0, 1>, TB, VEX_4V;
2700 defm V#NAME#PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2701 !strconcat(OpcodeStr, "pd"), f128mem,
2702 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2703 (bc_v2i64 (v2f64 VR128:$src2))))],
2704 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2705 (memopv2i64 addr:$src2)))], 0>,
2707 let Constraints = "$src1 = $dst" in {
2708 defm PS : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedSingle,
2709 !strconcat(OpcodeStr, "ps"), f128mem,
2710 [(set VR128:$dst, (v2i64 (OpNode VR128:$src1, VR128:$src2)))],
2711 [(set VR128:$dst, (OpNode (bc_v2i64 (v4f32 VR128:$src1)),
2712 (memopv2i64 addr:$src2)))]>, TB;
2714 defm PD : sse12_fp_packed_logical_rm<opc, VR128, SSEPackedDouble,
2715 !strconcat(OpcodeStr, "pd"), f128mem,
2716 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2717 (bc_v2i64 (v2f64 VR128:$src2))))],
2718 [(set VR128:$dst, (OpNode (bc_v2i64 (v2f64 VR128:$src1)),
2719 (memopv2i64 addr:$src2)))]>, TB, OpSize;
2723 /// sse12_fp_packed_logical_y - AVX 256-bit SSE 1 & 2 logical ops forms
2725 multiclass sse12_fp_packed_logical_y<bits<8> opc, string OpcodeStr,
2727 defm PSY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedSingle,
2728 !strconcat(OpcodeStr, "ps"), f256mem,
2729 [(set VR256:$dst, (v4i64 (OpNode VR256:$src1, VR256:$src2)))],
2730 [(set VR256:$dst, (OpNode (bc_v4i64 (v8f32 VR256:$src1)),
2731 (memopv4i64 addr:$src2)))], 0>, TB, VEX_4V;
2733 defm PDY : sse12_fp_packed_logical_rm<opc, VR256, SSEPackedDouble,
2734 !strconcat(OpcodeStr, "pd"), f256mem,
2735 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2736 (bc_v4i64 (v4f64 VR256:$src2))))],
2737 [(set VR256:$dst, (OpNode (bc_v4i64 (v4f64 VR256:$src1)),
2738 (memopv4i64 addr:$src2)))], 0>,
2742 // AVX 256-bit packed logical ops forms
2743 defm VAND : sse12_fp_packed_logical_y<0x54, "and", and>;
2744 defm VOR : sse12_fp_packed_logical_y<0x56, "or", or>;
2745 defm VXOR : sse12_fp_packed_logical_y<0x57, "xor", xor>;
2746 defm VANDN : sse12_fp_packed_logical_y<0x55, "andn", X86andnp>;
2748 defm AND : sse12_fp_packed_logical<0x54, "and", and>;
2749 defm OR : sse12_fp_packed_logical<0x56, "or", or>;
2750 defm XOR : sse12_fp_packed_logical<0x57, "xor", xor>;
2751 let isCommutable = 0 in
2752 defm ANDN : sse12_fp_packed_logical<0x55, "andn", X86andnp>;
2754 //===----------------------------------------------------------------------===//
2755 // SSE 1 & 2 - Arithmetic Instructions
2756 //===----------------------------------------------------------------------===//
2758 /// basic_sse12_fp_binop_xxx - SSE 1 & 2 binops come in both scalar and
2761 /// In addition, we also have a special variant of the scalar form here to
2762 /// represent the associated intrinsic operation. This form is unlike the
2763 /// plain scalar form, in that it takes an entire vector (instead of a scalar)
2764 /// and leaves the top elements unmodified (therefore these cannot be commuted).
2766 /// These three forms can each be reg+reg or reg+mem.
2769 /// FIXME: once all 256-bit intrinsics are matched, cleanup and refactor those
2771 multiclass basic_sse12_fp_binop_s<bits<8> opc, string OpcodeStr, SDNode OpNode,
2773 defm SS : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "ss"),
2774 OpNode, FR32, f32mem, Is2Addr>, XS;
2775 defm SD : sse12_fp_scalar<opc, !strconcat(OpcodeStr, "sd"),
2776 OpNode, FR64, f64mem, Is2Addr>, XD;
2779 multiclass basic_sse12_fp_binop_p<bits<8> opc, string OpcodeStr, SDNode OpNode,
2781 let mayLoad = 0 in {
2782 defm PS : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR128,
2783 v4f32, f128mem, memopv4f32, SSEPackedSingle, Is2Addr>, TB;
2784 defm PD : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR128,
2785 v2f64, f128mem, memopv2f64, SSEPackedDouble, Is2Addr>, TB, OpSize;
2789 multiclass basic_sse12_fp_binop_p_y<bits<8> opc, string OpcodeStr,
2791 let mayLoad = 0 in {
2792 defm PSY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "ps"), OpNode, VR256,
2793 v8f32, f256mem, memopv8f32, SSEPackedSingle, 0>, TB;
2794 defm PDY : sse12_fp_packed<opc, !strconcat(OpcodeStr, "pd"), OpNode, VR256,
2795 v4f64, f256mem, memopv4f64, SSEPackedDouble, 0>, TB, OpSize;
2799 multiclass basic_sse12_fp_binop_s_int<bits<8> opc, string OpcodeStr,
2801 defm SS : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2802 !strconcat(OpcodeStr, "ss"), "", "_ss", ssmem, sse_load_f32, Is2Addr>, XS;
2803 defm SD : sse12_fp_scalar_int<opc, OpcodeStr, VR128,
2804 !strconcat(OpcodeStr, "sd"), "2", "_sd", sdmem, sse_load_f64, Is2Addr>, XD;
2807 multiclass basic_sse12_fp_binop_p_int<bits<8> opc, string OpcodeStr,
2809 defm PS : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2810 !strconcat(OpcodeStr, "ps"), "sse", "_ps", f128mem, memopv4f32,
2811 SSEPackedSingle, Is2Addr>, TB;
2813 defm PD : sse12_fp_packed_int<opc, OpcodeStr, VR128,
2814 !strconcat(OpcodeStr, "pd"), "sse2", "_pd", f128mem, memopv2f64,
2815 SSEPackedDouble, Is2Addr>, TB, OpSize;
2818 multiclass basic_sse12_fp_binop_p_y_int<bits<8> opc, string OpcodeStr> {
2819 defm PSY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2820 !strconcat(OpcodeStr, "ps"), "avx", "_ps_256", f256mem, memopv8f32,
2821 SSEPackedSingle, 0>, TB;
2823 defm PDY : sse12_fp_packed_int<opc, OpcodeStr, VR256,
2824 !strconcat(OpcodeStr, "pd"), "avx", "_pd_256", f256mem, memopv4f64,
2825 SSEPackedDouble, 0>, TB, OpSize;
2828 // Binary Arithmetic instructions
2829 defm VADD : basic_sse12_fp_binop_s<0x58, "add", fadd, 0>,
2830 basic_sse12_fp_binop_s_int<0x58, "add", 0>, VEX_4V, VEX_LIG;
2831 defm VADD : basic_sse12_fp_binop_p<0x58, "add", fadd, 0>,
2832 basic_sse12_fp_binop_p_y<0x58, "add", fadd>, VEX_4V;
2833 defm VMUL : basic_sse12_fp_binop_s<0x59, "mul", fmul, 0>,
2834 basic_sse12_fp_binop_s_int<0x59, "mul", 0>, VEX_4V, VEX_LIG;
2835 defm VMUL : basic_sse12_fp_binop_p<0x59, "mul", fmul, 0>,
2836 basic_sse12_fp_binop_p_y<0x59, "mul", fmul>, VEX_4V;
2838 let isCommutable = 0 in {
2839 defm VSUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub, 0>,
2840 basic_sse12_fp_binop_s_int<0x5C, "sub", 0>, VEX_4V, VEX_LIG;
2841 defm VSUB : basic_sse12_fp_binop_p<0x5C, "sub", fsub, 0>,
2842 basic_sse12_fp_binop_p_y<0x5C, "sub", fsub>, VEX_4V;
2843 defm VDIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv, 0>,
2844 basic_sse12_fp_binop_s_int<0x5E, "div", 0>, VEX_4V, VEX_LIG;
2845 defm VDIV : basic_sse12_fp_binop_p<0x5E, "div", fdiv, 0>,
2846 basic_sse12_fp_binop_p_y<0x5E, "div", fdiv>, VEX_4V;
2847 defm VMAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax, 0>,
2848 basic_sse12_fp_binop_s_int<0x5F, "max", 0>, VEX_4V, VEX_LIG;
2849 defm VMAX : basic_sse12_fp_binop_p<0x5F, "max", X86fmax, 0>,
2850 basic_sse12_fp_binop_p_int<0x5F, "max", 0>,
2851 basic_sse12_fp_binop_p_y<0x5F, "max", X86fmax>,
2852 basic_sse12_fp_binop_p_y_int<0x5F, "max">, VEX_4V;
2853 defm VMIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin, 0>,
2854 basic_sse12_fp_binop_s_int<0x5D, "min", 0>, VEX_4V, VEX_LIG;
2855 defm VMIN : basic_sse12_fp_binop_p<0x5D, "min", X86fmin, 0>,
2856 basic_sse12_fp_binop_p_int<0x5D, "min", 0>,
2857 basic_sse12_fp_binop_p_y_int<0x5D, "min">,
2858 basic_sse12_fp_binop_p_y<0x5D, "min", X86fmin>, VEX_4V;
2861 let Constraints = "$src1 = $dst" in {
2862 defm ADD : basic_sse12_fp_binop_s<0x58, "add", fadd>,
2863 basic_sse12_fp_binop_p<0x58, "add", fadd>,
2864 basic_sse12_fp_binop_s_int<0x58, "add">;
2865 defm MUL : basic_sse12_fp_binop_s<0x59, "mul", fmul>,
2866 basic_sse12_fp_binop_p<0x59, "mul", fmul>,
2867 basic_sse12_fp_binop_s_int<0x59, "mul">;
2869 let isCommutable = 0 in {
2870 defm SUB : basic_sse12_fp_binop_s<0x5C, "sub", fsub>,
2871 basic_sse12_fp_binop_p<0x5C, "sub", fsub>,
2872 basic_sse12_fp_binop_s_int<0x5C, "sub">;
2873 defm DIV : basic_sse12_fp_binop_s<0x5E, "div", fdiv>,
2874 basic_sse12_fp_binop_p<0x5E, "div", fdiv>,
2875 basic_sse12_fp_binop_s_int<0x5E, "div">;
2876 defm MAX : basic_sse12_fp_binop_s<0x5F, "max", X86fmax>,
2877 basic_sse12_fp_binop_p<0x5F, "max", X86fmax>,
2878 basic_sse12_fp_binop_s_int<0x5F, "max">,
2879 basic_sse12_fp_binop_p_int<0x5F, "max">;
2880 defm MIN : basic_sse12_fp_binop_s<0x5D, "min", X86fmin>,
2881 basic_sse12_fp_binop_p<0x5D, "min", X86fmin>,
2882 basic_sse12_fp_binop_s_int<0x5D, "min">,
2883 basic_sse12_fp_binop_p_int<0x5D, "min">;
2888 /// In addition, we also have a special variant of the scalar form here to
2889 /// represent the associated intrinsic operation. This form is unlike the
2890 /// plain scalar form, in that it takes an entire vector (instead of a
2891 /// scalar) and leaves the top elements undefined.
2893 /// And, we have a special variant form for a full-vector intrinsic form.
2895 /// sse1_fp_unop_s - SSE1 unops in scalar form.
2896 multiclass sse1_fp_unop_s<bits<8> opc, string OpcodeStr,
2897 SDNode OpNode, Intrinsic F32Int> {
2898 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src),
2899 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2900 [(set FR32:$dst, (OpNode FR32:$src))]>;
2901 // For scalar unary operations, fold a load into the operation
2902 // only in OptForSize mode. It eliminates an instruction, but it also
2903 // eliminates a whole-register clobber (the load), so it introduces a
2904 // partial register update condition.
2905 def SSm : I<opc, MRMSrcMem, (outs FR32:$dst), (ins f32mem:$src),
2906 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2907 [(set FR32:$dst, (OpNode (load addr:$src)))]>, XS,
2908 Requires<[HasSSE1, OptForSize]>;
2909 def SSr_Int : SSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2910 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2911 [(set VR128:$dst, (F32Int VR128:$src))]>;
2912 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst), (ins ssmem:$src),
2913 !strconcat(OpcodeStr, "ss\t{$src, $dst|$dst, $src}"),
2914 [(set VR128:$dst, (F32Int sse_load_f32:$src))]>;
2917 /// sse1_fp_unop_s_avx - AVX SSE1 unops in scalar form.
2918 multiclass sse1_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2919 def SSr : SSI<opc, MRMSrcReg, (outs FR32:$dst), (ins FR32:$src1, FR32:$src2),
2920 !strconcat(OpcodeStr,
2921 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2923 def SSm : SSI<opc, MRMSrcMem, (outs FR32:$dst), (ins FR32:$src1,f32mem:$src2),
2924 !strconcat(OpcodeStr,
2925 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2926 def SSm_Int : SSI<opc, MRMSrcMem, (outs VR128:$dst),
2927 (ins VR128:$src1, ssmem:$src2),
2928 !strconcat(OpcodeStr,
2929 "ss\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
2932 /// sse1_fp_unop_p - SSE1 unops in packed form.
2933 multiclass sse1_fp_unop_p<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2934 def PSr : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2935 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2936 [(set VR128:$dst, (v4f32 (OpNode VR128:$src)))]>;
2937 def PSm : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2938 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2939 [(set VR128:$dst, (OpNode (memopv4f32 addr:$src)))]>;
2942 /// sse1_fp_unop_p_y - AVX 256-bit SSE1 unops in packed form.
2943 multiclass sse1_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
2944 def PSYr : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2945 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2946 [(set VR256:$dst, (v8f32 (OpNode VR256:$src)))]>;
2947 def PSYm : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2948 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2949 [(set VR256:$dst, (OpNode (memopv8f32 addr:$src)))]>;
2952 /// sse1_fp_unop_p_int - SSE1 intrinsics unops in packed forms.
2953 multiclass sse1_fp_unop_p_int<bits<8> opc, string OpcodeStr,
2954 Intrinsic V4F32Int> {
2955 def PSr_Int : PSI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2956 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2957 [(set VR128:$dst, (V4F32Int VR128:$src))]>;
2958 def PSm_Int : PSI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
2959 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2960 [(set VR128:$dst, (V4F32Int (memopv4f32 addr:$src)))]>;
2963 /// sse1_fp_unop_p_y_int - AVX 256-bit intrinsics unops in packed forms.
2964 multiclass sse1_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
2965 Intrinsic V4F32Int> {
2966 def PSYr_Int : PSI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
2967 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2968 [(set VR256:$dst, (V4F32Int VR256:$src))]>;
2969 def PSYm_Int : PSI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
2970 !strconcat(OpcodeStr, "ps\t{$src, $dst|$dst, $src}"),
2971 [(set VR256:$dst, (V4F32Int (memopv8f32 addr:$src)))]>;
2974 /// sse2_fp_unop_s - SSE2 unops in scalar form.
2975 multiclass sse2_fp_unop_s<bits<8> opc, string OpcodeStr,
2976 SDNode OpNode, Intrinsic F64Int> {
2977 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src),
2978 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2979 [(set FR64:$dst, (OpNode FR64:$src))]>;
2980 // See the comments in sse1_fp_unop_s for why this is OptForSize.
2981 def SDm : I<opc, MRMSrcMem, (outs FR64:$dst), (ins f64mem:$src),
2982 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2983 [(set FR64:$dst, (OpNode (load addr:$src)))]>, XD,
2984 Requires<[HasSSE2, OptForSize]>;
2985 def SDr_Int : SDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
2986 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2987 [(set VR128:$dst, (F64Int VR128:$src))]>;
2988 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst), (ins sdmem:$src),
2989 !strconcat(OpcodeStr, "sd\t{$src, $dst|$dst, $src}"),
2990 [(set VR128:$dst, (F64Int sse_load_f64:$src))]>;
2993 /// sse2_fp_unop_s_avx - AVX SSE2 unops in scalar form.
2994 multiclass sse2_fp_unop_s_avx<bits<8> opc, string OpcodeStr> {
2995 let neverHasSideEffects = 1 in {
2996 def SDr : SDI<opc, MRMSrcReg, (outs FR64:$dst), (ins FR64:$src1, FR64:$src2),
2997 !strconcat(OpcodeStr,
2998 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3000 def SDm : SDI<opc, MRMSrcMem, (outs FR64:$dst), (ins FR64:$src1,f64mem:$src2),
3001 !strconcat(OpcodeStr,
3002 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3004 def SDm_Int : SDI<opc, MRMSrcMem, (outs VR128:$dst),
3005 (ins VR128:$src1, sdmem:$src2),
3006 !strconcat(OpcodeStr,
3007 "sd\t{$src2, $src1, $dst|$dst, $src1, $src2}"), []>;
3010 /// sse2_fp_unop_p - SSE2 unops in vector forms.
3011 multiclass sse2_fp_unop_p<bits<8> opc, string OpcodeStr,
3013 def PDr : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3014 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3015 [(set VR128:$dst, (v2f64 (OpNode VR128:$src)))]>;
3016 def PDm : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3017 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3018 [(set VR128:$dst, (OpNode (memopv2f64 addr:$src)))]>;
3021 /// sse2_fp_unop_p_y - AVX SSE2 256-bit unops in vector forms.
3022 multiclass sse2_fp_unop_p_y<bits<8> opc, string OpcodeStr, SDNode OpNode> {
3023 def PDYr : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3024 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3025 [(set VR256:$dst, (v4f64 (OpNode VR256:$src)))]>;
3026 def PDYm : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3027 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3028 [(set VR256:$dst, (OpNode (memopv4f64 addr:$src)))]>;
3031 /// sse2_fp_unop_p_int - SSE2 intrinsic unops in vector forms.
3032 multiclass sse2_fp_unop_p_int<bits<8> opc, string OpcodeStr,
3033 Intrinsic V2F64Int> {
3034 def PDr_Int : PDI<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3035 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3036 [(set VR128:$dst, (V2F64Int VR128:$src))]>;
3037 def PDm_Int : PDI<opc, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
3038 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3039 [(set VR128:$dst, (V2F64Int (memopv2f64 addr:$src)))]>;
3042 /// sse2_fp_unop_p_y_int - AVX 256-bit intrinsic unops in vector forms.
3043 multiclass sse2_fp_unop_p_y_int<bits<8> opc, string OpcodeStr,
3044 Intrinsic V2F64Int> {
3045 def PDYr_Int : PDI<opc, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3046 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3047 [(set VR256:$dst, (V2F64Int VR256:$src))]>;
3048 def PDYm_Int : PDI<opc, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
3049 !strconcat(OpcodeStr, "pd\t{$src, $dst|$dst, $src}"),
3050 [(set VR256:$dst, (V2F64Int (memopv4f64 addr:$src)))]>;
3053 let Predicates = [HasAVX] in {
3055 defm VSQRT : sse1_fp_unop_s_avx<0x51, "vsqrt">,
3056 sse2_fp_unop_s_avx<0x51, "vsqrt">, VEX_4V, VEX_LIG;
3058 defm VSQRT : sse1_fp_unop_p<0x51, "vsqrt", fsqrt>,
3059 sse2_fp_unop_p<0x51, "vsqrt", fsqrt>,
3060 sse1_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3061 sse2_fp_unop_p_y<0x51, "vsqrt", fsqrt>,
3062 sse1_fp_unop_p_int<0x51, "vsqrt", int_x86_sse_sqrt_ps>,
3063 sse2_fp_unop_p_int<0x51, "vsqrt", int_x86_sse2_sqrt_pd>,
3064 sse1_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_ps_256>,
3065 sse2_fp_unop_p_y_int<0x51, "vsqrt", int_x86_avx_sqrt_pd_256>,
3068 // Reciprocal approximations. Note that these typically require refinement
3069 // in order to obtain suitable precision.
3070 defm VRSQRT : sse1_fp_unop_s_avx<0x52, "vrsqrt">, VEX_4V, VEX_LIG;
3071 defm VRSQRT : sse1_fp_unop_p<0x52, "vrsqrt", X86frsqrt>,
3072 sse1_fp_unop_p_y<0x52, "vrsqrt", X86frsqrt>,
3073 sse1_fp_unop_p_y_int<0x52, "vrsqrt", int_x86_avx_rsqrt_ps_256>,
3074 sse1_fp_unop_p_int<0x52, "vrsqrt", int_x86_sse_rsqrt_ps>, VEX;
3076 defm VRCP : sse1_fp_unop_s_avx<0x53, "vrcp">, VEX_4V, VEX_LIG;
3077 defm VRCP : sse1_fp_unop_p<0x53, "vrcp", X86frcp>,
3078 sse1_fp_unop_p_y<0x53, "vrcp", X86frcp>,
3079 sse1_fp_unop_p_y_int<0x53, "vrcp", int_x86_avx_rcp_ps_256>,
3080 sse1_fp_unop_p_int<0x53, "vrcp", int_x86_sse_rcp_ps>, VEX;
3083 def : Pat<(f32 (fsqrt FR32:$src)),
3084 (VSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3085 def : Pat<(f32 (fsqrt (load addr:$src))),
3086 (VSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3087 Requires<[HasAVX, OptForSize]>;
3088 def : Pat<(f64 (fsqrt FR64:$src)),
3089 (VSQRTSDr (f64 (IMPLICIT_DEF)), FR64:$src)>, Requires<[HasAVX]>;
3090 def : Pat<(f64 (fsqrt (load addr:$src))),
3091 (VSQRTSDm (f64 (IMPLICIT_DEF)), addr:$src)>,
3092 Requires<[HasAVX, OptForSize]>;
3094 def : Pat<(f32 (X86frsqrt FR32:$src)),
3095 (VRSQRTSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3096 def : Pat<(f32 (X86frsqrt (load addr:$src))),
3097 (VRSQRTSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3098 Requires<[HasAVX, OptForSize]>;
3100 def : Pat<(f32 (X86frcp FR32:$src)),
3101 (VRCPSSr (f32 (IMPLICIT_DEF)), FR32:$src)>, Requires<[HasAVX]>;
3102 def : Pat<(f32 (X86frcp (load addr:$src))),
3103 (VRCPSSm (f32 (IMPLICIT_DEF)), addr:$src)>,
3104 Requires<[HasAVX, OptForSize]>;
3106 let Predicates = [HasAVX] in {
3107 def : Pat<(int_x86_sse_sqrt_ss VR128:$src),
3108 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3109 (VSQRTSSr (f32 (IMPLICIT_DEF)),
3110 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3112 def : Pat<(int_x86_sse_sqrt_ss sse_load_f32:$src),
3113 (VSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3115 def : Pat<(int_x86_sse2_sqrt_sd VR128:$src),
3116 (INSERT_SUBREG (v2f64 (IMPLICIT_DEF)),
3117 (VSQRTSDr (f64 (IMPLICIT_DEF)),
3118 (EXTRACT_SUBREG (v2f64 VR128:$src), sub_sd)),
3120 def : Pat<(int_x86_sse2_sqrt_sd sse_load_f64:$src),
3121 (VSQRTSDm_Int (v2f64 (IMPLICIT_DEF)), sse_load_f64:$src)>;
3123 def : Pat<(int_x86_sse_rsqrt_ss VR128:$src),
3124 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3125 (VRSQRTSSr (f32 (IMPLICIT_DEF)),
3126 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3128 def : Pat<(int_x86_sse_rsqrt_ss sse_load_f32:$src),
3129 (VRSQRTSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3131 def : Pat<(int_x86_sse_rcp_ss VR128:$src),
3132 (INSERT_SUBREG (v4f32 (IMPLICIT_DEF)),
3133 (VRCPSSr (f32 (IMPLICIT_DEF)),
3134 (EXTRACT_SUBREG (v4f32 VR128:$src), sub_ss)),
3136 def : Pat<(int_x86_sse_rcp_ss sse_load_f32:$src),
3137 (VRCPSSm_Int (v4f32 (IMPLICIT_DEF)), sse_load_f32:$src)>;
3141 defm SQRT : sse1_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse_sqrt_ss>,
3142 sse1_fp_unop_p<0x51, "sqrt", fsqrt>,
3143 sse1_fp_unop_p_int<0x51, "sqrt", int_x86_sse_sqrt_ps>,
3144 sse2_fp_unop_s<0x51, "sqrt", fsqrt, int_x86_sse2_sqrt_sd>,
3145 sse2_fp_unop_p<0x51, "sqrt", fsqrt>,
3146 sse2_fp_unop_p_int<0x51, "sqrt", int_x86_sse2_sqrt_pd>;
3148 // Reciprocal approximations. Note that these typically require refinement
3149 // in order to obtain suitable precision.
3150 defm RSQRT : sse1_fp_unop_s<0x52, "rsqrt", X86frsqrt, int_x86_sse_rsqrt_ss>,
3151 sse1_fp_unop_p<0x52, "rsqrt", X86frsqrt>,
3152 sse1_fp_unop_p_int<0x52, "rsqrt", int_x86_sse_rsqrt_ps>;
3153 defm RCP : sse1_fp_unop_s<0x53, "rcp", X86frcp, int_x86_sse_rcp_ss>,
3154 sse1_fp_unop_p<0x53, "rcp", X86frcp>,
3155 sse1_fp_unop_p_int<0x53, "rcp", int_x86_sse_rcp_ps>;
3157 // There is no f64 version of the reciprocal approximation instructions.
3159 //===----------------------------------------------------------------------===//
3160 // SSE 1 & 2 - Non-temporal stores
3161 //===----------------------------------------------------------------------===//
3163 let AddedComplexity = 400 in { // Prefer non-temporal versions
3164 def VMOVNTPSmr : VPSI<0x2B, MRMDestMem, (outs),
3165 (ins f128mem:$dst, VR128:$src),
3166 "movntps\t{$src, $dst|$dst, $src}",
3167 [(alignednontemporalstore (v4f32 VR128:$src),
3169 def VMOVNTPDmr : VPDI<0x2B, MRMDestMem, (outs),
3170 (ins f128mem:$dst, VR128:$src),
3171 "movntpd\t{$src, $dst|$dst, $src}",
3172 [(alignednontemporalstore (v2f64 VR128:$src),
3174 def VMOVNTDQ_64mr : VPDI<0xE7, MRMDestMem, (outs),
3175 (ins f128mem:$dst, VR128:$src),
3176 "movntdq\t{$src, $dst|$dst, $src}",
3177 [(alignednontemporalstore (v2f64 VR128:$src),
3180 let ExeDomain = SSEPackedInt in
3181 def VMOVNTDQmr : VPDI<0xE7, MRMDestMem, (outs),
3182 (ins f128mem:$dst, VR128:$src),
3183 "movntdq\t{$src, $dst|$dst, $src}",
3184 [(alignednontemporalstore (v4f32 VR128:$src),
3187 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3188 (VMOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasAVX]>;
3190 def VMOVNTPSYmr : VPSI<0x2B, MRMDestMem, (outs),
3191 (ins f256mem:$dst, VR256:$src),
3192 "movntps\t{$src, $dst|$dst, $src}",
3193 [(alignednontemporalstore (v8f32 VR256:$src),
3195 def VMOVNTPDYmr : VPDI<0x2B, MRMDestMem, (outs),
3196 (ins f256mem:$dst, VR256:$src),
3197 "movntpd\t{$src, $dst|$dst, $src}",
3198 [(alignednontemporalstore (v4f64 VR256:$src),
3200 def VMOVNTDQY_64mr : VPDI<0xE7, MRMDestMem, (outs),
3201 (ins f256mem:$dst, VR256:$src),
3202 "movntdq\t{$src, $dst|$dst, $src}",
3203 [(alignednontemporalstore (v4f64 VR256:$src),
3205 let ExeDomain = SSEPackedInt in
3206 def VMOVNTDQYmr : VPDI<0xE7, MRMDestMem, (outs),
3207 (ins f256mem:$dst, VR256:$src),
3208 "movntdq\t{$src, $dst|$dst, $src}",
3209 [(alignednontemporalstore (v8f32 VR256:$src),
3213 def : Pat<(int_x86_avx_movnt_dq_256 addr:$dst, VR256:$src),
3214 (VMOVNTDQYmr addr:$dst, VR256:$src)>;
3215 def : Pat<(int_x86_avx_movnt_pd_256 addr:$dst, VR256:$src),
3216 (VMOVNTPDYmr addr:$dst, VR256:$src)>;
3217 def : Pat<(int_x86_avx_movnt_ps_256 addr:$dst, VR256:$src),
3218 (VMOVNTPSYmr addr:$dst, VR256:$src)>;
3220 let AddedComplexity = 400 in { // Prefer non-temporal versions
3221 def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3222 "movntps\t{$src, $dst|$dst, $src}",
3223 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3224 def MOVNTPDmr : PDI<0x2B, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3225 "movntpd\t{$src, $dst|$dst, $src}",
3226 [(alignednontemporalstore(v2f64 VR128:$src), addr:$dst)]>;
3228 def MOVNTDQ_64mr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3229 "movntdq\t{$src, $dst|$dst, $src}",
3230 [(alignednontemporalstore (v2f64 VR128:$src), addr:$dst)]>;
3232 let ExeDomain = SSEPackedInt in
3233 def MOVNTDQmr : PDI<0xE7, MRMDestMem, (outs), (ins f128mem:$dst, VR128:$src),
3234 "movntdq\t{$src, $dst|$dst, $src}",
3235 [(alignednontemporalstore (v4f32 VR128:$src), addr:$dst)]>;
3237 def : Pat<(alignednontemporalstore (v2i64 VR128:$src), addr:$dst),
3238 (MOVNTDQmr addr:$dst, VR128:$src)>, Requires<[HasSSE2]>;
3240 // There is no AVX form for instructions below this point
3241 def MOVNTImr : I<0xC3, MRMDestMem, (outs), (ins i32mem:$dst, GR32:$src),
3242 "movnti{l}\t{$src, $dst|$dst, $src}",
3243 [(nontemporalstore (i32 GR32:$src), addr:$dst)]>,
3244 TB, Requires<[HasSSE2]>;
3245 def MOVNTI_64mr : RI<0xC3, MRMDestMem, (outs), (ins i64mem:$dst, GR64:$src),
3246 "movnti{q}\t{$src, $dst|$dst, $src}",
3247 [(nontemporalstore (i64 GR64:$src), addr:$dst)]>,
3248 TB, Requires<[HasSSE2]>;
3251 //===----------------------------------------------------------------------===//
3252 // SSE 1 & 2 - Prefetch and memory fence
3253 //===----------------------------------------------------------------------===//
3255 // Prefetch intrinsic.
3256 def PREFETCHT0 : VoPSI<0x18, MRM1m, (outs), (ins i8mem:$src),
3257 "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3), (i32 1))]>;
3258 def PREFETCHT1 : VoPSI<0x18, MRM2m, (outs), (ins i8mem:$src),
3259 "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2), (i32 1))]>;
3260 def PREFETCHT2 : VoPSI<0x18, MRM3m, (outs), (ins i8mem:$src),
3261 "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1), (i32 1))]>;
3262 def PREFETCHNTA : VoPSI<0x18, MRM0m, (outs), (ins i8mem:$src),
3263 "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0), (i32 1))]>;
3266 def CLFLUSH : I<0xAE, MRM7m, (outs), (ins i8mem:$src),
3267 "clflush\t$src", [(int_x86_sse2_clflush addr:$src)]>,
3268 TB, Requires<[HasSSE2]>;
3270 // Pause. This "instruction" is encoded as "rep; nop", so even though it
3271 // was introduced with SSE2, it's backward compatible.
3272 def PAUSE : I<0x90, RawFrm, (outs), (ins), "pause", []>, REP;
3274 // Load, store, and memory fence
3275 def SFENCE : I<0xAE, MRM_F8, (outs), (ins),
3276 "sfence", [(int_x86_sse_sfence)]>, TB, Requires<[HasSSE1]>;
3277 def LFENCE : I<0xAE, MRM_E8, (outs), (ins),
3278 "lfence", [(int_x86_sse2_lfence)]>, TB, Requires<[HasSSE2]>;
3279 def MFENCE : I<0xAE, MRM_F0, (outs), (ins),
3280 "mfence", [(int_x86_sse2_mfence)]>, TB, Requires<[HasSSE2]>;
3282 def : Pat<(X86SFence), (SFENCE)>;
3283 def : Pat<(X86LFence), (LFENCE)>;
3284 def : Pat<(X86MFence), (MFENCE)>;
3286 //===----------------------------------------------------------------------===//
3287 // SSE 1 & 2 - Load/Store XCSR register
3288 //===----------------------------------------------------------------------===//
3290 def VLDMXCSR : VPSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3291 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>, VEX;
3292 def VSTMXCSR : VPSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3293 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>, VEX;
3295 def LDMXCSR : PSI<0xAE, MRM2m, (outs), (ins i32mem:$src),
3296 "ldmxcsr\t$src", [(int_x86_sse_ldmxcsr addr:$src)]>;
3297 def STMXCSR : PSI<0xAE, MRM3m, (outs), (ins i32mem:$dst),
3298 "stmxcsr\t$dst", [(int_x86_sse_stmxcsr addr:$dst)]>;
3300 //===---------------------------------------------------------------------===//
3301 // SSE2 - Move Aligned/Unaligned Packed Integer Instructions
3302 //===---------------------------------------------------------------------===//
3304 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3306 let neverHasSideEffects = 1 in {
3307 def VMOVDQArr : VPDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3308 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3309 def VMOVDQAYrr : VPDI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3310 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3312 def VMOVDQUrr : VSSI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3313 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3314 def VMOVDQUYrr : VSSI<0x6F, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
3315 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3318 let isCodeGenOnly = 1 in {
3319 def VMOVDQArr_REV : VPDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3320 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3321 def VMOVDQAYrr_REV : VPDI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3322 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3323 def VMOVDQUrr_REV : VSSI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3324 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3325 def VMOVDQUYrr_REV : VSSI<0x7F, MRMDestReg, (outs VR256:$dst), (ins VR256:$src),
3326 "movdqu\t{$src, $dst|$dst, $src}", []>, VEX;
3329 let canFoldAsLoad = 1, mayLoad = 1 in {
3330 def VMOVDQArm : VPDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3331 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3332 def VMOVDQAYrm : VPDI<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3333 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3334 let Predicates = [HasAVX] in {
3335 def VMOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3336 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3337 def VMOVDQUYrm : I<0x6F, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
3338 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3342 let mayStore = 1 in {
3343 def VMOVDQAmr : VPDI<0x7F, MRMDestMem, (outs),
3344 (ins i128mem:$dst, VR128:$src),
3345 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3346 def VMOVDQAYmr : VPDI<0x7F, MRMDestMem, (outs),
3347 (ins i256mem:$dst, VR256:$src),
3348 "movdqa\t{$src, $dst|$dst, $src}", []>, VEX;
3349 let Predicates = [HasAVX] in {
3350 def VMOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3351 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3352 def VMOVDQUYmr : I<0x7F, MRMDestMem, (outs), (ins i256mem:$dst, VR256:$src),
3353 "vmovdqu\t{$src, $dst|$dst, $src}",[]>, XS, VEX;
3357 let neverHasSideEffects = 1 in
3358 def MOVDQArr : PDI<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3359 "movdqa\t{$src, $dst|$dst, $src}", []>;
3361 def MOVDQUrr : I<0x6F, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
3362 "movdqu\t{$src, $dst|$dst, $src}",
3363 []>, XS, Requires<[HasSSE2]>;
3366 let isCodeGenOnly = 1 in {
3367 def MOVDQArr_REV : PDI<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3368 "movdqa\t{$src, $dst|$dst, $src}", []>;
3370 def MOVDQUrr_REV : I<0x7F, MRMDestReg, (outs VR128:$dst), (ins VR128:$src),
3371 "movdqu\t{$src, $dst|$dst, $src}",
3372 []>, XS, Requires<[HasSSE2]>;
3375 let canFoldAsLoad = 1, mayLoad = 1 in {
3376 def MOVDQArm : PDI<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3377 "movdqa\t{$src, $dst|$dst, $src}",
3378 [/*(set VR128:$dst, (alignedloadv2i64 addr:$src))*/]>;
3379 def MOVDQUrm : I<0x6F, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
3380 "movdqu\t{$src, $dst|$dst, $src}",
3381 [/*(set VR128:$dst, (loadv2i64 addr:$src))*/]>,
3382 XS, Requires<[HasSSE2]>;
3385 let mayStore = 1 in {
3386 def MOVDQAmr : PDI<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3387 "movdqa\t{$src, $dst|$dst, $src}",
3388 [/*(alignedstore (v2i64 VR128:$src), addr:$dst)*/]>;
3389 def MOVDQUmr : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3390 "movdqu\t{$src, $dst|$dst, $src}",
3391 [/*(store (v2i64 VR128:$src), addr:$dst)*/]>,
3392 XS, Requires<[HasSSE2]>;
3395 // Intrinsic forms of MOVDQU load and store
3396 def VMOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3397 "vmovdqu\t{$src, $dst|$dst, $src}",
3398 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3399 XS, VEX, Requires<[HasAVX]>;
3401 def MOVDQUmr_Int : I<0x7F, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
3402 "movdqu\t{$src, $dst|$dst, $src}",
3403 [(int_x86_sse2_storeu_dq addr:$dst, VR128:$src)]>,
3404 XS, Requires<[HasSSE2]>;
3406 } // ExeDomain = SSEPackedInt
3408 let Predicates = [HasAVX] in {
3409 def : Pat<(int_x86_avx_loadu_dq_256 addr:$src), (VMOVDQUYrm addr:$src)>;
3410 def : Pat<(int_x86_avx_storeu_dq_256 addr:$dst, VR256:$src),
3411 (VMOVDQUYmr addr:$dst, VR256:$src)>;
3414 //===---------------------------------------------------------------------===//
3415 // SSE2 - Packed Integer Arithmetic Instructions
3416 //===---------------------------------------------------------------------===//
3418 let ExeDomain = SSEPackedInt in { // SSE integer instructions
3420 multiclass PDI_binop_rm_int<bits<8> opc, string OpcodeStr, Intrinsic IntId,
3421 RegisterClass RC, PatFrag memop_frag,
3422 X86MemOperand x86memop, bit IsCommutable = 0,
3424 let isCommutable = IsCommutable in
3425 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3426 (ins RC:$src1, RC:$src2),
3428 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3429 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3430 [(set RC:$dst, (IntId RC:$src1, RC:$src2))]>;
3431 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3432 (ins RC:$src1, x86memop:$src2),
3434 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3435 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3436 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memop_frag addr:$src2))))]>;
3439 multiclass PDI_binop_rmi_int<bits<8> opc, bits<8> opc2, Format ImmForm,
3440 string OpcodeStr, Intrinsic IntId,
3441 Intrinsic IntId2, RegisterClass RC,
3443 // src2 is always 128-bit
3444 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3445 (ins RC:$src1, VR128:$src2),
3447 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3448 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3449 [(set RC:$dst, (IntId RC:$src1, VR128:$src2))]>;
3450 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3451 (ins RC:$src1, i128mem:$src2),
3453 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3454 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3455 [(set RC:$dst, (IntId RC:$src1, (bitconvert (memopv2i64 addr:$src2))))]>;
3456 def ri : PDIi8<opc2, ImmForm, (outs RC:$dst),
3457 (ins RC:$src1, i32i8imm:$src2),
3459 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3460 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3461 [(set RC:$dst, (IntId2 RC:$src1, (i32 imm:$src2)))]>;
3464 /// PDI_binop_rm - Simple SSE2 binary operator.
3465 multiclass PDI_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
3466 ValueType OpVT, RegisterClass RC, PatFrag memop_frag,
3467 X86MemOperand x86memop, bit IsCommutable = 0,
3469 let isCommutable = IsCommutable in
3470 def rr : PDI<opc, MRMSrcReg, (outs RC:$dst),
3471 (ins RC:$src1, RC:$src2),
3473 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3474 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3475 [(set RC:$dst, (OpVT (OpNode RC:$src1, RC:$src2)))]>;
3476 def rm : PDI<opc, MRMSrcMem, (outs RC:$dst),
3477 (ins RC:$src1, x86memop:$src2),
3479 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
3480 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
3481 [(set RC:$dst, (OpVT (OpNode RC:$src1,
3482 (bitconvert (memop_frag addr:$src2)))))]>;
3484 } // ExeDomain = SSEPackedInt
3486 // 128-bit Integer Arithmetic
3488 let Predicates = [HasAVX] in {
3489 defm VPADDB : PDI_binop_rm<0xFC, "vpaddb", add, v16i8, VR128, memopv2i64,
3490 i128mem, 1, 0 /*3addr*/>, VEX_4V;
3491 defm VPADDW : PDI_binop_rm<0xFD, "vpaddw", add, v8i16, VR128, memopv2i64,
3492 i128mem, 1, 0>, VEX_4V;
3493 defm VPADDD : PDI_binop_rm<0xFE, "vpaddd", add, v4i32, VR128, memopv2i64,
3494 i128mem, 1, 0>, VEX_4V;
3495 defm VPADDQ : PDI_binop_rm<0xD4, "vpaddq", add, v2i64, VR128, memopv2i64,
3496 i128mem, 1, 0>, VEX_4V;
3497 defm VPMULLW : PDI_binop_rm<0xD5, "vpmullw", mul, v8i16, VR128, memopv2i64,
3498 i128mem, 1, 0>, VEX_4V;
3499 defm VPSUBB : PDI_binop_rm<0xF8, "vpsubb", sub, v16i8, VR128, memopv2i64,
3500 i128mem, 0, 0>, VEX_4V;
3501 defm VPSUBW : PDI_binop_rm<0xF9, "vpsubw", sub, v8i16, VR128, memopv2i64,
3502 i128mem, 0, 0>, VEX_4V;
3503 defm VPSUBD : PDI_binop_rm<0xFA, "vpsubd", sub, v4i32, VR128, memopv2i64,
3504 i128mem, 0, 0>, VEX_4V;
3505 defm VPSUBQ : PDI_binop_rm<0xFB, "vpsubq", sub, v2i64, VR128, memopv2i64,
3506 i128mem, 0, 0>, VEX_4V;
3509 defm VPSUBSB : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_sse2_psubs_b,
3510 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3511 defm VPSUBSW : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_sse2_psubs_w,
3512 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3513 defm VPSUBUSB : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_sse2_psubus_b,
3514 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3515 defm VPSUBUSW : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_sse2_psubus_w,
3516 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3517 defm VPADDSB : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_sse2_padds_b,
3518 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3519 defm VPADDSW : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_sse2_padds_w,
3520 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3521 defm VPADDUSB : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_sse2_paddus_b,
3522 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3523 defm VPADDUSW : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_sse2_paddus_w,
3524 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3525 defm VPMULHUW : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_sse2_pmulhu_w,
3526 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3527 defm VPMULHW : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_sse2_pmulh_w,
3528 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3529 defm VPMULUDQ : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_sse2_pmulu_dq,
3530 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3531 defm VPMADDWD : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_sse2_pmadd_wd,
3532 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3533 defm VPAVGB : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_sse2_pavg_b,
3534 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3535 defm VPAVGW : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_sse2_pavg_w,
3536 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3537 defm VPMINUB : PDI_binop_rm_int<0xDA, "vpminub", int_x86_sse2_pminu_b,
3538 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3539 defm VPMINSW : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_sse2_pmins_w,
3540 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3541 defm VPMAXUB : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_sse2_pmaxu_b,
3542 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3543 defm VPMAXSW : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_sse2_pmaxs_w,
3544 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3545 defm VPSADBW : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_sse2_psad_bw,
3546 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3549 let Predicates = [HasAVX2] in {
3550 defm VPADDBY : PDI_binop_rm<0xFC, "vpaddb", add, v32i8, VR256, memopv4i64,
3551 i256mem, 1, 0>, VEX_4V;
3552 defm VPADDWY : PDI_binop_rm<0xFD, "vpaddw", add, v16i16, VR256, memopv4i64,
3553 i256mem, 1, 0>, VEX_4V;
3554 defm VPADDDY : PDI_binop_rm<0xFE, "vpaddd", add, v8i32, VR256, memopv4i64,
3555 i256mem, 1, 0>, VEX_4V;
3556 defm VPADDQY : PDI_binop_rm<0xD4, "vpaddq", add, v4i64, VR256, memopv4i64,
3557 i256mem, 1, 0>, VEX_4V;
3558 defm VPMULLWY : PDI_binop_rm<0xD5, "vpmullw", mul, v16i16, VR256, memopv4i64,
3559 i256mem, 1, 0>, VEX_4V;
3560 defm VPSUBBY : PDI_binop_rm<0xF8, "vpsubb", sub, v32i8, VR256, memopv4i64,
3561 i256mem, 0, 0>, VEX_4V;
3562 defm VPSUBWY : PDI_binop_rm<0xF9, "vpsubw", sub, v16i16,VR256, memopv4i64,
3563 i256mem, 0, 0>, VEX_4V;
3564 defm VPSUBDY : PDI_binop_rm<0xFA, "vpsubd", sub, v8i32, VR256, memopv4i64,
3565 i256mem, 0, 0>, VEX_4V;
3566 defm VPSUBQY : PDI_binop_rm<0xFB, "vpsubq", sub, v4i64, VR256, memopv4i64,
3567 i256mem, 0, 0>, VEX_4V;
3570 defm VPSUBSBY : PDI_binop_rm_int<0xE8, "vpsubsb" , int_x86_avx2_psubs_b,
3571 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3572 defm VPSUBSWY : PDI_binop_rm_int<0xE9, "vpsubsw" , int_x86_avx2_psubs_w,
3573 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3574 defm VPSUBUSBY : PDI_binop_rm_int<0xD8, "vpsubusb", int_x86_avx2_psubus_b,
3575 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3576 defm VPSUBUSWY : PDI_binop_rm_int<0xD9, "vpsubusw", int_x86_avx2_psubus_w,
3577 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3578 defm VPADDSBY : PDI_binop_rm_int<0xEC, "vpaddsb" , int_x86_avx2_padds_b,
3579 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3580 defm VPADDSWY : PDI_binop_rm_int<0xED, "vpaddsw" , int_x86_avx2_padds_w,
3581 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3582 defm VPADDUSBY : PDI_binop_rm_int<0xDC, "vpaddusb", int_x86_avx2_paddus_b,
3583 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3584 defm VPADDUSWY : PDI_binop_rm_int<0xDD, "vpaddusw", int_x86_avx2_paddus_w,
3585 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3586 defm VPMULHUWY : PDI_binop_rm_int<0xE4, "vpmulhuw", int_x86_avx2_pmulhu_w,
3587 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3588 defm VPMULHWY : PDI_binop_rm_int<0xE5, "vpmulhw" , int_x86_avx2_pmulh_w,
3589 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3590 defm VPMULUDQY : PDI_binop_rm_int<0xF4, "vpmuludq", int_x86_avx2_pmulu_dq,
3591 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3592 defm VPMADDWDY : PDI_binop_rm_int<0xF5, "vpmaddwd", int_x86_avx2_pmadd_wd,
3593 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3594 defm VPAVGBY : PDI_binop_rm_int<0xE0, "vpavgb", int_x86_avx2_pavg_b,
3595 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3596 defm VPAVGWY : PDI_binop_rm_int<0xE3, "vpavgw", int_x86_avx2_pavg_w,
3597 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3598 defm VPMINUBY : PDI_binop_rm_int<0xDA, "vpminub", int_x86_avx2_pminu_b,
3599 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3600 defm VPMINSWY : PDI_binop_rm_int<0xEA, "vpminsw", int_x86_avx2_pmins_w,
3601 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3602 defm VPMAXUBY : PDI_binop_rm_int<0xDE, "vpmaxub", int_x86_avx2_pmaxu_b,
3603 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3604 defm VPMAXSWY : PDI_binop_rm_int<0xEE, "vpmaxsw", int_x86_avx2_pmaxs_w,
3605 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3606 defm VPSADBWY : PDI_binop_rm_int<0xF6, "vpsadbw", int_x86_avx2_psad_bw,
3607 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3610 let Constraints = "$src1 = $dst" in {
3611 defm PADDB : PDI_binop_rm<0xFC, "paddb", add, v16i8, VR128, memopv2i64,
3613 defm PADDW : PDI_binop_rm<0xFD, "paddw", add, v8i16, VR128, memopv2i64,
3615 defm PADDD : PDI_binop_rm<0xFE, "paddd", add, v4i32, VR128, memopv2i64,
3617 defm PADDQ : PDI_binop_rm<0xD4, "paddq", add, v2i64, VR128, memopv2i64,
3619 defm PMULLW : PDI_binop_rm<0xD5, "pmullw", mul, v8i16, VR128, memopv2i64,
3621 defm PSUBB : PDI_binop_rm<0xF8, "psubb", sub, v16i8, VR128, memopv2i64,
3623 defm PSUBW : PDI_binop_rm<0xF9, "psubw", sub, v8i16, VR128, memopv2i64,
3625 defm PSUBD : PDI_binop_rm<0xFA, "psubd", sub, v4i32, VR128, memopv2i64,
3627 defm PSUBQ : PDI_binop_rm<0xFB, "psubq", sub, v2i64, VR128, memopv2i64,
3631 defm PSUBSB : PDI_binop_rm_int<0xE8, "psubsb" , int_x86_sse2_psubs_b,
3632 VR128, memopv2i64, i128mem>;
3633 defm PSUBSW : PDI_binop_rm_int<0xE9, "psubsw" , int_x86_sse2_psubs_w,
3634 VR128, memopv2i64, i128mem>;
3635 defm PSUBUSB : PDI_binop_rm_int<0xD8, "psubusb", int_x86_sse2_psubus_b,
3636 VR128, memopv2i64, i128mem>;
3637 defm PSUBUSW : PDI_binop_rm_int<0xD9, "psubusw", int_x86_sse2_psubus_w,
3638 VR128, memopv2i64, i128mem>;
3639 defm PADDSB : PDI_binop_rm_int<0xEC, "paddsb" , int_x86_sse2_padds_b,
3640 VR128, memopv2i64, i128mem, 1>;
3641 defm PADDSW : PDI_binop_rm_int<0xED, "paddsw" , int_x86_sse2_padds_w,
3642 VR128, memopv2i64, i128mem, 1>;
3643 defm PADDUSB : PDI_binop_rm_int<0xDC, "paddusb", int_x86_sse2_paddus_b,
3644 VR128, memopv2i64, i128mem, 1>;
3645 defm PADDUSW : PDI_binop_rm_int<0xDD, "paddusw", int_x86_sse2_paddus_w,
3646 VR128, memopv2i64, i128mem, 1>;
3647 defm PMULHUW : PDI_binop_rm_int<0xE4, "pmulhuw", int_x86_sse2_pmulhu_w,
3648 VR128, memopv2i64, i128mem, 1>;
3649 defm PMULHW : PDI_binop_rm_int<0xE5, "pmulhw" , int_x86_sse2_pmulh_w,
3650 VR128, memopv2i64, i128mem, 1>;
3651 defm PMULUDQ : PDI_binop_rm_int<0xF4, "pmuludq", int_x86_sse2_pmulu_dq,
3652 VR128, memopv2i64, i128mem, 1>;
3653 defm PMADDWD : PDI_binop_rm_int<0xF5, "pmaddwd", int_x86_sse2_pmadd_wd,
3654 VR128, memopv2i64, i128mem, 1>;
3655 defm PAVGB : PDI_binop_rm_int<0xE0, "pavgb", int_x86_sse2_pavg_b,
3656 VR128, memopv2i64, i128mem, 1>;
3657 defm PAVGW : PDI_binop_rm_int<0xE3, "pavgw", int_x86_sse2_pavg_w,
3658 VR128, memopv2i64, i128mem, 1>;
3659 defm PMINUB : PDI_binop_rm_int<0xDA, "pminub", int_x86_sse2_pminu_b,
3660 VR128, memopv2i64, i128mem, 1>;
3661 defm PMINSW : PDI_binop_rm_int<0xEA, "pminsw", int_x86_sse2_pmins_w,
3662 VR128, memopv2i64, i128mem, 1>;
3663 defm PMAXUB : PDI_binop_rm_int<0xDE, "pmaxub", int_x86_sse2_pmaxu_b,
3664 VR128, memopv2i64, i128mem, 1>;
3665 defm PMAXSW : PDI_binop_rm_int<0xEE, "pmaxsw", int_x86_sse2_pmaxs_w,
3666 VR128, memopv2i64, i128mem, 1>;
3667 defm PSADBW : PDI_binop_rm_int<0xF6, "psadbw", int_x86_sse2_psad_bw,
3668 VR128, memopv2i64, i128mem, 1>;
3670 } // Constraints = "$src1 = $dst"
3672 //===---------------------------------------------------------------------===//
3673 // SSE2 - Packed Integer Logical Instructions
3674 //===---------------------------------------------------------------------===//
3676 let Predicates = [HasAVX] in {
3677 defm VPSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3678 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3680 defm VPSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3681 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3683 defm VPSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3684 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3687 defm VPSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3688 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3690 defm VPSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3691 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3693 defm VPSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3694 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3697 defm VPSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3698 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3700 defm VPSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3701 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3704 defm VPAND : PDI_binop_rm<0xDB, "vpand", and, v2i64, VR128, memopv2i64,
3705 i128mem, 1, 0>, VEX_4V;
3706 defm VPOR : PDI_binop_rm<0xEB, "vpor" , or, v2i64, VR128, memopv2i64,
3707 i128mem, 1, 0>, VEX_4V;
3708 defm VPXOR : PDI_binop_rm<0xEF, "vpxor", xor, v2i64, VR128, memopv2i64,
3709 i128mem, 1, 0>, VEX_4V;
3710 defm VPANDN : PDI_binop_rm<0xDF, "vpandn", X86andnp, v2i64, VR128, memopv2i64,
3711 i128mem, 0, 0>, VEX_4V;
3713 let ExeDomain = SSEPackedInt in {
3714 let neverHasSideEffects = 1 in {
3715 // 128-bit logical shifts.
3716 def VPSLLDQri : PDIi8<0x73, MRM7r,
3717 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3718 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3720 def VPSRLDQri : PDIi8<0x73, MRM3r,
3721 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3722 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3724 // PSRADQri doesn't exist in SSE[1-3].
3729 let Predicates = [HasAVX2] in {
3730 defm VPSLLWY : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "vpsllw",
3731 int_x86_avx2_psll_w, int_x86_avx2_pslli_w,
3733 defm VPSLLDY : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "vpslld",
3734 int_x86_avx2_psll_d, int_x86_avx2_pslli_d,
3736 defm VPSLLQY : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "vpsllq",
3737 int_x86_avx2_psll_q, int_x86_avx2_pslli_q,
3740 defm VPSRLWY : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "vpsrlw",
3741 int_x86_avx2_psrl_w, int_x86_avx2_psrli_w,
3743 defm VPSRLDY : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "vpsrld",
3744 int_x86_avx2_psrl_d, int_x86_avx2_psrli_d,
3746 defm VPSRLQY : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "vpsrlq",
3747 int_x86_avx2_psrl_q, int_x86_avx2_psrli_q,
3750 defm VPSRAWY : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "vpsraw",
3751 int_x86_avx2_psra_w, int_x86_avx2_psrai_w,
3753 defm VPSRADY : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "vpsrad",
3754 int_x86_avx2_psra_d, int_x86_avx2_psrai_d,
3757 defm VPANDY : PDI_binop_rm<0xDB, "vpand", and, v4i64, VR256, memopv4i64,
3758 i256mem, 1, 0>, VEX_4V;
3759 defm VPORY : PDI_binop_rm<0xEB, "vpor", or, v4i64, VR256, memopv4i64,
3760 i256mem, 1, 0>, VEX_4V;
3761 defm VPXORY : PDI_binop_rm<0xEF, "vpxor", xor, v4i64, VR256, memopv4i64,
3762 i256mem, 1, 0>, VEX_4V;
3763 defm VPANDNY : PDI_binop_rm<0xDF, "vpandn", X86andnp, v4i64, VR256, memopv4i64,
3764 i256mem, 0, 0>, VEX_4V;
3766 let ExeDomain = SSEPackedInt in {
3767 let neverHasSideEffects = 1 in {
3768 // 128-bit logical shifts.
3769 def VPSLLDQYri : PDIi8<0x73, MRM7r,
3770 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3771 "vpslldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3773 def VPSRLDQYri : PDIi8<0x73, MRM3r,
3774 (outs VR256:$dst), (ins VR256:$src1, i32i8imm:$src2),
3775 "vpsrldq\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
3777 // PSRADQYri doesn't exist in SSE[1-3].
3782 let Constraints = "$src1 = $dst" in {
3783 defm PSLLW : PDI_binop_rmi_int<0xF1, 0x71, MRM6r, "psllw",
3784 int_x86_sse2_psll_w, int_x86_sse2_pslli_w,
3786 defm PSLLD : PDI_binop_rmi_int<0xF2, 0x72, MRM6r, "pslld",
3787 int_x86_sse2_psll_d, int_x86_sse2_pslli_d,
3789 defm PSLLQ : PDI_binop_rmi_int<0xF3, 0x73, MRM6r, "psllq",
3790 int_x86_sse2_psll_q, int_x86_sse2_pslli_q,
3793 defm PSRLW : PDI_binop_rmi_int<0xD1, 0x71, MRM2r, "psrlw",
3794 int_x86_sse2_psrl_w, int_x86_sse2_psrli_w,
3796 defm PSRLD : PDI_binop_rmi_int<0xD2, 0x72, MRM2r, "psrld",
3797 int_x86_sse2_psrl_d, int_x86_sse2_psrli_d,
3799 defm PSRLQ : PDI_binop_rmi_int<0xD3, 0x73, MRM2r, "psrlq",
3800 int_x86_sse2_psrl_q, int_x86_sse2_psrli_q,
3803 defm PSRAW : PDI_binop_rmi_int<0xE1, 0x71, MRM4r, "psraw",
3804 int_x86_sse2_psra_w, int_x86_sse2_psrai_w,
3806 defm PSRAD : PDI_binop_rmi_int<0xE2, 0x72, MRM4r, "psrad",
3807 int_x86_sse2_psra_d, int_x86_sse2_psrai_d,
3810 defm PAND : PDI_binop_rm<0xDB, "pand", and, v2i64, VR128, memopv2i64,
3812 defm POR : PDI_binop_rm<0xEB, "por" , or, v2i64, VR128, memopv2i64,
3814 defm PXOR : PDI_binop_rm<0xEF, "pxor", xor, v2i64, VR128, memopv2i64,
3816 defm PANDN : PDI_binop_rm<0xDF, "pandn", X86andnp, v2i64, VR128, memopv2i64,
3819 let ExeDomain = SSEPackedInt in {
3820 let neverHasSideEffects = 1 in {
3821 // 128-bit logical shifts.
3822 def PSLLDQri : PDIi8<0x73, MRM7r,
3823 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3824 "pslldq\t{$src2, $dst|$dst, $src2}", []>;
3825 def PSRLDQri : PDIi8<0x73, MRM3r,
3826 (outs VR128:$dst), (ins VR128:$src1, i32i8imm:$src2),
3827 "psrldq\t{$src2, $dst|$dst, $src2}", []>;
3828 // PSRADQri doesn't exist in SSE[1-3].
3831 } // Constraints = "$src1 = $dst"
3833 let Predicates = [HasAVX] in {
3834 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3835 (VPSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3836 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3837 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3838 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3839 (VPSLLDQri VR128:$src1, imm:$src2)>;
3840 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3841 (VPSRLDQri VR128:$src1, imm:$src2)>;
3842 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3843 (VPSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3845 // Shift up / down and insert zero's.
3846 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3847 (VPSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3848 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3849 (VPSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3852 let Predicates = [HasAVX2] in {
3853 def : Pat<(int_x86_avx2_psll_dq VR256:$src1, imm:$src2),
3854 (VPSLLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3855 def : Pat<(int_x86_avx2_psrl_dq VR256:$src1, imm:$src2),
3856 (VPSRLDQYri VR256:$src1, (BYTE_imm imm:$src2))>;
3857 def : Pat<(int_x86_avx2_psll_dq_bs VR256:$src1, imm:$src2),
3858 (VPSLLDQYri VR256:$src1, imm:$src2)>;
3859 def : Pat<(int_x86_avx2_psrl_dq_bs VR256:$src1, imm:$src2),
3860 (VPSRLDQYri VR256:$src1, imm:$src2)>;
3863 let Predicates = [HasSSE2] in {
3864 def : Pat<(int_x86_sse2_psll_dq VR128:$src1, imm:$src2),
3865 (PSLLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3866 def : Pat<(int_x86_sse2_psrl_dq VR128:$src1, imm:$src2),
3867 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3868 def : Pat<(int_x86_sse2_psll_dq_bs VR128:$src1, imm:$src2),
3869 (PSLLDQri VR128:$src1, imm:$src2)>;
3870 def : Pat<(int_x86_sse2_psrl_dq_bs VR128:$src1, imm:$src2),
3871 (PSRLDQri VR128:$src1, imm:$src2)>;
3872 def : Pat<(v2f64 (X86fsrl VR128:$src1, i32immSExt8:$src2)),
3873 (PSRLDQri VR128:$src1, (BYTE_imm imm:$src2))>;
3875 // Shift up / down and insert zero's.
3876 def : Pat<(v2i64 (X86vshl VR128:$src, (i8 imm:$amt))),
3877 (PSLLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3878 def : Pat<(v2i64 (X86vshr VR128:$src, (i8 imm:$amt))),
3879 (PSRLDQri VR128:$src, (BYTE_imm imm:$amt))>;
3882 //===---------------------------------------------------------------------===//
3883 // SSE2 - Packed Integer Comparison Instructions
3884 //===---------------------------------------------------------------------===//
3886 let Predicates = [HasAVX] in {
3887 defm VPCMPEQB : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_sse2_pcmpeq_b,
3888 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3889 defm VPCMPEQW : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_sse2_pcmpeq_w,
3890 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3891 defm VPCMPEQD : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_sse2_pcmpeq_d,
3892 VR128, memopv2i64, i128mem, 1, 0>, VEX_4V;
3893 defm VPCMPGTB : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_sse2_pcmpgt_b,
3894 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3895 defm VPCMPGTW : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_sse2_pcmpgt_w,
3896 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3897 defm VPCMPGTD : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_sse2_pcmpgt_d,
3898 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
3900 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3901 (VPCMPEQBrr VR128:$src1, VR128:$src2)>;
3902 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3903 (bc_v16i8 (memopv2i64 addr:$src2)))),
3904 (VPCMPEQBrm VR128:$src1, addr:$src2)>;
3905 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
3906 (VPCMPEQWrr VR128:$src1, VR128:$src2)>;
3907 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
3908 (bc_v8i16 (memopv2i64 addr:$src2)))),
3909 (VPCMPEQWrm VR128:$src1, addr:$src2)>;
3910 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
3911 (VPCMPEQDrr VR128:$src1, VR128:$src2)>;
3912 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
3913 (bc_v4i32 (memopv2i64 addr:$src2)))),
3914 (VPCMPEQDrm VR128:$src1, addr:$src2)>;
3916 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
3917 (VPCMPGTBrr VR128:$src1, VR128:$src2)>;
3918 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
3919 (bc_v16i8 (memopv2i64 addr:$src2)))),
3920 (VPCMPGTBrm VR128:$src1, addr:$src2)>;
3921 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
3922 (VPCMPGTWrr VR128:$src1, VR128:$src2)>;
3923 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
3924 (bc_v8i16 (memopv2i64 addr:$src2)))),
3925 (VPCMPGTWrm VR128:$src1, addr:$src2)>;
3926 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
3927 (VPCMPGTDrr VR128:$src1, VR128:$src2)>;
3928 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
3929 (bc_v4i32 (memopv2i64 addr:$src2)))),
3930 (VPCMPGTDrm VR128:$src1, addr:$src2)>;
3933 let Predicates = [HasAVX2] in {
3934 defm VPCMPEQBY : PDI_binop_rm_int<0x74, "vpcmpeqb", int_x86_avx2_pcmpeq_b,
3935 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3936 defm VPCMPEQWY : PDI_binop_rm_int<0x75, "vpcmpeqw", int_x86_avx2_pcmpeq_w,
3937 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3938 defm VPCMPEQDY : PDI_binop_rm_int<0x76, "vpcmpeqd", int_x86_avx2_pcmpeq_d,
3939 VR256, memopv4i64, i256mem, 1, 0>, VEX_4V;
3940 defm VPCMPGTBY : PDI_binop_rm_int<0x64, "vpcmpgtb", int_x86_avx2_pcmpgt_b,
3941 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3942 defm VPCMPGTWY : PDI_binop_rm_int<0x65, "vpcmpgtw", int_x86_avx2_pcmpgt_w,
3943 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3944 defm VPCMPGTDY : PDI_binop_rm_int<0x66, "vpcmpgtd", int_x86_avx2_pcmpgt_d,
3945 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
3947 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1, VR256:$src2)),
3948 (VPCMPEQBYrr VR256:$src1, VR256:$src2)>;
3949 def : Pat<(v32i8 (X86pcmpeqb VR256:$src1,
3950 (bc_v32i8 (memopv4i64 addr:$src2)))),
3951 (VPCMPEQBYrm VR256:$src1, addr:$src2)>;
3952 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1, VR256:$src2)),
3953 (VPCMPEQWYrr VR256:$src1, VR256:$src2)>;
3954 def : Pat<(v16i16 (X86pcmpeqw VR256:$src1,
3955 (bc_v16i16 (memopv4i64 addr:$src2)))),
3956 (VPCMPEQWYrm VR256:$src1, addr:$src2)>;
3957 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1, VR256:$src2)),
3958 (VPCMPEQDYrr VR256:$src1, VR256:$src2)>;
3959 def : Pat<(v8i32 (X86pcmpeqd VR256:$src1,
3960 (bc_v8i32 (memopv4i64 addr:$src2)))),
3961 (VPCMPEQDYrm VR256:$src1, addr:$src2)>;
3963 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1, VR256:$src2)),
3964 (VPCMPGTBYrr VR256:$src1, VR256:$src2)>;
3965 def : Pat<(v32i8 (X86pcmpgtb VR256:$src1,
3966 (bc_v32i8 (memopv4i64 addr:$src2)))),
3967 (VPCMPGTBYrm VR256:$src1, addr:$src2)>;
3968 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1, VR256:$src2)),
3969 (VPCMPGTWYrr VR256:$src1, VR256:$src2)>;
3970 def : Pat<(v16i16 (X86pcmpgtw VR256:$src1,
3971 (bc_v16i16 (memopv4i64 addr:$src2)))),
3972 (VPCMPGTWYrm VR256:$src1, addr:$src2)>;
3973 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1, VR256:$src2)),
3974 (VPCMPGTDYrr VR256:$src1, VR256:$src2)>;
3975 def : Pat<(v8i32 (X86pcmpgtd VR256:$src1,
3976 (bc_v8i32 (memopv4i64 addr:$src2)))),
3977 (VPCMPGTDYrm VR256:$src1, addr:$src2)>;
3980 let Constraints = "$src1 = $dst" in {
3981 defm PCMPEQB : PDI_binop_rm_int<0x74, "pcmpeqb", int_x86_sse2_pcmpeq_b,
3982 VR128, memopv2i64, i128mem, 1>;
3983 defm PCMPEQW : PDI_binop_rm_int<0x75, "pcmpeqw", int_x86_sse2_pcmpeq_w,
3984 VR128, memopv2i64, i128mem, 1>;
3985 defm PCMPEQD : PDI_binop_rm_int<0x76, "pcmpeqd", int_x86_sse2_pcmpeq_d,
3986 VR128, memopv2i64, i128mem, 1>;
3987 defm PCMPGTB : PDI_binop_rm_int<0x64, "pcmpgtb", int_x86_sse2_pcmpgt_b,
3988 VR128, memopv2i64, i128mem>;
3989 defm PCMPGTW : PDI_binop_rm_int<0x65, "pcmpgtw", int_x86_sse2_pcmpgt_w,
3990 VR128, memopv2i64, i128mem>;
3991 defm PCMPGTD : PDI_binop_rm_int<0x66, "pcmpgtd", int_x86_sse2_pcmpgt_d,
3992 VR128, memopv2i64, i128mem>;
3993 } // Constraints = "$src1 = $dst"
3995 let Predicates = [HasSSE2] in {
3996 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1, VR128:$src2)),
3997 (PCMPEQBrr VR128:$src1, VR128:$src2)>;
3998 def : Pat<(v16i8 (X86pcmpeqb VR128:$src1,
3999 (bc_v16i8 (memopv2i64 addr:$src2)))),
4000 (PCMPEQBrm VR128:$src1, addr:$src2)>;
4001 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1, VR128:$src2)),
4002 (PCMPEQWrr VR128:$src1, VR128:$src2)>;
4003 def : Pat<(v8i16 (X86pcmpeqw VR128:$src1,
4004 (bc_v8i16 (memopv2i64 addr:$src2)))),
4005 (PCMPEQWrm VR128:$src1, addr:$src2)>;
4006 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1, VR128:$src2)),
4007 (PCMPEQDrr VR128:$src1, VR128:$src2)>;
4008 def : Pat<(v4i32 (X86pcmpeqd VR128:$src1,
4009 (bc_v4i32 (memopv2i64 addr:$src2)))),
4010 (PCMPEQDrm VR128:$src1, addr:$src2)>;
4012 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1, VR128:$src2)),
4013 (PCMPGTBrr VR128:$src1, VR128:$src2)>;
4014 def : Pat<(v16i8 (X86pcmpgtb VR128:$src1,
4015 (bc_v16i8 (memopv2i64 addr:$src2)))),
4016 (PCMPGTBrm VR128:$src1, addr:$src2)>;
4017 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1, VR128:$src2)),
4018 (PCMPGTWrr VR128:$src1, VR128:$src2)>;
4019 def : Pat<(v8i16 (X86pcmpgtw VR128:$src1,
4020 (bc_v8i16 (memopv2i64 addr:$src2)))),
4021 (PCMPGTWrm VR128:$src1, addr:$src2)>;
4022 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1, VR128:$src2)),
4023 (PCMPGTDrr VR128:$src1, VR128:$src2)>;
4024 def : Pat<(v4i32 (X86pcmpgtd VR128:$src1,
4025 (bc_v4i32 (memopv2i64 addr:$src2)))),
4026 (PCMPGTDrm VR128:$src1, addr:$src2)>;
4029 //===---------------------------------------------------------------------===//
4030 // SSE2 - Packed Integer Pack Instructions
4031 //===---------------------------------------------------------------------===//
4033 let Predicates = [HasAVX] in {
4034 defm VPACKSSWB : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_sse2_packsswb_128,
4035 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4036 defm VPACKSSDW : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_sse2_packssdw_128,
4037 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4038 defm VPACKUSWB : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_sse2_packuswb_128,
4039 VR128, memopv2i64, i128mem, 0, 0>, VEX_4V;
4042 let Predicates = [HasAVX2] in {
4043 defm VPACKSSWBY : PDI_binop_rm_int<0x63, "vpacksswb", int_x86_avx2_packsswb,
4044 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4045 defm VPACKSSDWY : PDI_binop_rm_int<0x6B, "vpackssdw", int_x86_avx2_packssdw,
4046 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4047 defm VPACKUSWBY : PDI_binop_rm_int<0x67, "vpackuswb", int_x86_avx2_packuswb,
4048 VR256, memopv4i64, i256mem, 0, 0>, VEX_4V;
4051 let Constraints = "$src1 = $dst" in {
4052 defm PACKSSWB : PDI_binop_rm_int<0x63, "packsswb", int_x86_sse2_packsswb_128,
4053 VR128, memopv2i64, i128mem>;
4054 defm PACKSSDW : PDI_binop_rm_int<0x6B, "packssdw", int_x86_sse2_packssdw_128,
4055 VR128, memopv2i64, i128mem>;
4056 defm PACKUSWB : PDI_binop_rm_int<0x67, "packuswb", int_x86_sse2_packuswb_128,
4057 VR128, memopv2i64, i128mem>;
4058 } // Constraints = "$src1 = $dst"
4060 //===---------------------------------------------------------------------===//
4061 // SSE2 - Packed Integer Shuffle Instructions
4062 //===---------------------------------------------------------------------===//
4064 let ExeDomain = SSEPackedInt in {
4065 multiclass sse2_pshuffle<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4067 def ri : Ii8<0x70, MRMSrcReg,
4068 (outs VR128:$dst), (ins VR128:$src1, i8imm:$src2),
4069 !strconcat(OpcodeStr,
4070 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4071 [(set VR128:$dst, (vt (pshuf_frag:$src2 VR128:$src1,
4073 def mi : Ii8<0x70, MRMSrcMem,
4074 (outs VR128:$dst), (ins i128mem:$src1, i8imm:$src2),
4075 !strconcat(OpcodeStr,
4076 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4077 [(set VR128:$dst, (vt (pshuf_frag:$src2
4078 (bc_frag (memopv2i64 addr:$src1)),
4082 multiclass sse2_pshuffle_y<string OpcodeStr, ValueType vt, PatFrag pshuf_frag,
4084 def Yri : Ii8<0x70, MRMSrcReg,
4085 (outs VR256:$dst), (ins VR256:$src1, i8imm:$src2),
4086 !strconcat(OpcodeStr,
4087 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4088 [(set VR256:$dst, (vt (pshuf_frag:$src2 VR256:$src1,
4090 def Ymi : Ii8<0x70, MRMSrcMem,
4091 (outs VR256:$dst), (ins i256mem:$src1, i8imm:$src2),
4092 !strconcat(OpcodeStr,
4093 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4094 [(set VR256:$dst, (vt (pshuf_frag:$src2
4095 (bc_frag (memopv4i64 addr:$src1)),
4098 } // ExeDomain = SSEPackedInt
4100 let Predicates = [HasAVX] in {
4101 let AddedComplexity = 5 in
4102 defm VPSHUFD : sse2_pshuffle<"vpshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize,
4105 // SSE2 with ImmT == Imm8 and XS prefix.
4106 defm VPSHUFHW : sse2_pshuffle<"vpshufhw", v8i16, pshufhw, bc_v8i16>, XS,
4109 // SSE2 with ImmT == Imm8 and XD prefix.
4110 defm VPSHUFLW : sse2_pshuffle<"vpshuflw", v8i16, pshuflw, bc_v8i16>, XD,
4113 let AddedComplexity = 5 in
4114 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4115 (VPSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4116 // Unary v4f32 shuffle with VPSHUF* in order to fold a load.
4117 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4118 (VPSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4120 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4122 (VPSHUFDmi addr:$src1, imm:$imm)>;
4123 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4125 (VPSHUFDmi addr:$src1, imm:$imm)>;
4126 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4127 (VPSHUFDri VR128:$src1, imm:$imm)>;
4128 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4129 (VPSHUFDri VR128:$src1, imm:$imm)>;
4130 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4131 (VPSHUFHWri VR128:$src, imm:$imm)>;
4132 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4134 (VPSHUFHWmi addr:$src, imm:$imm)>;
4135 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4136 (VPSHUFLWri VR128:$src, imm:$imm)>;
4137 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4139 (VPSHUFLWmi addr:$src, imm:$imm)>;
4142 let Predicates = [HasAVX2] in {
4143 let AddedComplexity = 5 in
4144 defm VPSHUFD : sse2_pshuffle_y<"vpshufd", v8i32, pshufd, bc_v8i32>, TB,
4147 // SSE2 with ImmT == Imm8 and XS prefix.
4148 defm VPSHUFHW : sse2_pshuffle_y<"vpshufhw", v16i16, pshufhw, bc_v16i16>, XS,
4151 // SSE2 with ImmT == Imm8 and XD prefix.
4152 defm VPSHUFLW : sse2_pshuffle_y<"vpshuflw", v16i16, pshuflw, bc_v16i16>, XD,
4156 let Predicates = [HasSSE2] in {
4157 let AddedComplexity = 5 in
4158 defm PSHUFD : sse2_pshuffle<"pshufd", v4i32, pshufd, bc_v4i32>, TB, OpSize;
4160 // SSE2 with ImmT == Imm8 and XS prefix.
4161 defm PSHUFHW : sse2_pshuffle<"pshufhw", v8i16, pshufhw, bc_v8i16>, XS;
4163 // SSE2 with ImmT == Imm8 and XD prefix.
4164 defm PSHUFLW : sse2_pshuffle<"pshuflw", v8i16, pshuflw, bc_v8i16>, XD;
4166 let AddedComplexity = 5 in
4167 def : Pat<(v4f32 (pshufd:$src2 VR128:$src1, (undef))),
4168 (PSHUFDri VR128:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4169 // Unary v4f32 shuffle with PSHUF* in order to fold a load.
4170 def : Pat<(pshufd:$src2 (bc_v4i32 (memopv4f32 addr:$src1)), (undef)),
4171 (PSHUFDmi addr:$src1, (SHUFFLE_get_shuf_imm VR128:$src2))>;
4173 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv2i64 addr:$src1)),
4175 (PSHUFDmi addr:$src1, imm:$imm)>;
4176 def : Pat<(v4i32 (X86PShufd (bc_v4i32 (memopv4f32 addr:$src1)),
4178 (PSHUFDmi addr:$src1, imm:$imm)>;
4179 def : Pat<(v4f32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4180 (PSHUFDri VR128:$src1, imm:$imm)>;
4181 def : Pat<(v4i32 (X86PShufd VR128:$src1, (i8 imm:$imm))),
4182 (PSHUFDri VR128:$src1, imm:$imm)>;
4183 def : Pat<(v8i16 (X86PShufhw VR128:$src, (i8 imm:$imm))),
4184 (PSHUFHWri VR128:$src, imm:$imm)>;
4185 def : Pat<(v8i16 (X86PShufhw (bc_v8i16 (memopv2i64 addr:$src)),
4187 (PSHUFHWmi addr:$src, imm:$imm)>;
4188 def : Pat<(v8i16 (X86PShuflw VR128:$src, (i8 imm:$imm))),
4189 (PSHUFLWri VR128:$src, imm:$imm)>;
4190 def : Pat<(v8i16 (X86PShuflw (bc_v8i16 (memopv2i64 addr:$src)),
4192 (PSHUFLWmi addr:$src, imm:$imm)>;
4195 //===---------------------------------------------------------------------===//
4196 // SSE2 - Packed Integer Unpack Instructions
4197 //===---------------------------------------------------------------------===//
4199 let ExeDomain = SSEPackedInt in {
4200 multiclass sse2_unpack<bits<8> opc, string OpcodeStr, ValueType vt,
4201 SDNode OpNode, PatFrag bc_frag, bit Is2Addr = 1> {
4202 def rr : PDI<opc, MRMSrcReg,
4203 (outs VR128:$dst), (ins VR128:$src1, VR128:$src2),
4205 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4206 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4207 [(set VR128:$dst, (vt (OpNode VR128:$src1, VR128:$src2)))]>;
4208 def rm : PDI<opc, MRMSrcMem,
4209 (outs VR128:$dst), (ins VR128:$src1, i128mem:$src2),
4211 !strconcat(OpcodeStr,"\t{$src2, $dst|$dst, $src2}"),
4212 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
4213 [(set VR128:$dst, (OpNode VR128:$src1,
4214 (bc_frag (memopv2i64
4218 multiclass sse2_unpack_y<bits<8> opc, string OpcodeStr, ValueType vt,
4219 SDNode OpNode, PatFrag bc_frag> {
4220 def Yrr : PDI<opc, MRMSrcReg,
4221 (outs VR256:$dst), (ins VR256:$src1, VR256:$src2),
4222 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4223 [(set VR256:$dst, (vt (OpNode VR256:$src1, VR256:$src2)))]>;
4224 def Yrm : PDI<opc, MRMSrcMem,
4225 (outs VR256:$dst), (ins VR256:$src1, i256mem:$src2),
4226 !strconcat(OpcodeStr,"\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
4227 [(set VR256:$dst, (OpNode VR256:$src1,
4228 (bc_frag (memopv4i64 addr:$src2))))]>;
4231 let Predicates = [HasAVX] in {
4232 defm VPUNPCKLBW : sse2_unpack<0x60, "vpunpcklbw", v16i8, X86Unpckl,
4233 bc_v16i8, 0>, VEX_4V;
4234 defm VPUNPCKLWD : sse2_unpack<0x61, "vpunpcklwd", v8i16, X86Unpckl,
4235 bc_v8i16, 0>, VEX_4V;
4236 defm VPUNPCKLDQ : sse2_unpack<0x62, "vpunpckldq", v4i32, X86Unpckl,
4237 bc_v4i32, 0>, VEX_4V;
4238 defm VPUNPCKLQDQ : sse2_unpack<0x6C, "vpunpcklqdq", v2i64, X86Unpckl,
4239 bc_v2i64, 0>, VEX_4V;
4241 defm VPUNPCKHBW : sse2_unpack<0x68, "vpunpckhbw", v16i8, X86Unpckh,
4242 bc_v16i8, 0>, VEX_4V;
4243 defm VPUNPCKHWD : sse2_unpack<0x69, "vpunpckhwd", v8i16, X86Unpckh,
4244 bc_v8i16, 0>, VEX_4V;
4245 defm VPUNPCKHDQ : sse2_unpack<0x6A, "vpunpckhdq", v4i32, X86Unpckh,
4246 bc_v4i32, 0>, VEX_4V;
4247 defm VPUNPCKHQDQ : sse2_unpack<0x6D, "vpunpckhqdq", v2i64, X86Unpckh,
4248 bc_v2i64, 0>, VEX_4V;
4251 let Predicates = [HasAVX2] in {
4252 defm VPUNPCKLBW : sse2_unpack_y<0x60, "vpunpcklbw", v32i8, X86Unpckl,
4254 defm VPUNPCKLWD : sse2_unpack_y<0x61, "vpunpcklwd", v16i16, X86Unpckl,
4256 defm VPUNPCKLDQ : sse2_unpack_y<0x62, "vpunpckldq", v8i32, X86Unpckl,
4258 defm VPUNPCKLQDQ : sse2_unpack_y<0x6C, "vpunpcklqdq", v4i64, X86Unpckl,
4261 defm VPUNPCKHBW : sse2_unpack_y<0x68, "vpunpckhbw", v32i8, X86Unpckh,
4263 defm VPUNPCKHWD : sse2_unpack_y<0x69, "vpunpckhwd", v16i16, X86Unpckh,
4265 defm VPUNPCKHDQ : sse2_unpack_y<0x6A, "vpunpckhdq", v8i32, X86Unpckh,
4267 defm VPUNPCKHQDQ : sse2_unpack_y<0x6D, "vpunpckhqdq", v4i64, X86Unpckh,
4271 let Constraints = "$src1 = $dst" in {
4272 defm PUNPCKLBW : sse2_unpack<0x60, "punpcklbw", v16i8, X86Unpckl,
4274 defm PUNPCKLWD : sse2_unpack<0x61, "punpcklwd", v8i16, X86Unpckl,
4276 defm PUNPCKLDQ : sse2_unpack<0x62, "punpckldq", v4i32, X86Unpckl,
4278 defm PUNPCKLQDQ : sse2_unpack<0x6C, "punpcklqdq", v2i64, X86Unpckl,
4281 defm PUNPCKHBW : sse2_unpack<0x68, "punpckhbw", v16i8, X86Unpckh,
4283 defm PUNPCKHWD : sse2_unpack<0x69, "punpckhwd", v8i16, X86Unpckh,
4285 defm PUNPCKHDQ : sse2_unpack<0x6A, "punpckhdq", v4i32, X86Unpckh,
4287 defm PUNPCKHQDQ : sse2_unpack<0x6D, "punpckhqdq", v2i64, X86Unpckh,
4290 } // ExeDomain = SSEPackedInt
4292 // Patterns for using AVX1 instructions with integer vectors
4293 // Here to give AVX2 priority
4294 let Predicates = [HasAVX] in {
4295 def : Pat<(v8i32 (X86Unpckl VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4296 (VUNPCKLPSYrm VR256:$src1, addr:$src2)>;
4297 def : Pat<(v8i32 (X86Unpckl VR256:$src1, VR256:$src2)),
4298 (VUNPCKLPSYrr VR256:$src1, VR256:$src2)>;
4299 def : Pat<(v8i32 (X86Unpckh VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)))),
4300 (VUNPCKHPSYrm VR256:$src1, addr:$src2)>;
4301 def : Pat<(v8i32 (X86Unpckh VR256:$src1, VR256:$src2)),
4302 (VUNPCKHPSYrr VR256:$src1, VR256:$src2)>;
4304 def : Pat<(v4i64 (X86Unpckl VR256:$src1, (memopv4i64 addr:$src2))),
4305 (VUNPCKLPDYrm VR256:$src1, addr:$src2)>;
4306 def : Pat<(v4i64 (X86Unpckl VR256:$src1, VR256:$src2)),
4307 (VUNPCKLPDYrr VR256:$src1, VR256:$src2)>;
4308 def : Pat<(v4i64 (X86Unpckh VR256:$src1, (memopv4i64 addr:$src2))),
4309 (VUNPCKHPDYrm VR256:$src1, addr:$src2)>;
4310 def : Pat<(v4i64 (X86Unpckh VR256:$src1, VR256:$src2)),
4311 (VUNPCKHPDYrr VR256:$src1, VR256:$src2)>;
4314 // Splat v2f64 / v2i64
4315 let AddedComplexity = 10 in {
4316 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4317 (PUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasSSE2]>;
4318 def : Pat<(splat_lo (v2i64 VR128:$src), (undef)),
4319 (VPUNPCKLQDQrr VR128:$src, VR128:$src)>, Requires<[HasAVX]>;
4322 //===---------------------------------------------------------------------===//
4323 // SSE2 - Packed Integer Extract and Insert
4324 //===---------------------------------------------------------------------===//
4326 let ExeDomain = SSEPackedInt in {
4327 multiclass sse2_pinsrw<bit Is2Addr = 1> {
4328 def rri : Ii8<0xC4, MRMSrcReg,
4329 (outs VR128:$dst), (ins VR128:$src1,
4330 GR32:$src2, i32i8imm:$src3),
4332 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4333 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4335 (X86pinsrw VR128:$src1, GR32:$src2, imm:$src3))]>;
4336 def rmi : Ii8<0xC4, MRMSrcMem,
4337 (outs VR128:$dst), (ins VR128:$src1,
4338 i16mem:$src2, i32i8imm:$src3),
4340 "pinsrw\t{$src3, $src2, $dst|$dst, $src2, $src3}",
4341 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
4343 (X86pinsrw VR128:$src1, (extloadi16 addr:$src2),
4348 let Predicates = [HasAVX] in
4349 def VPEXTRWri : Ii8<0xC5, MRMSrcReg,
4350 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4351 "vpextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4352 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4353 imm:$src2))]>, TB, OpSize, VEX;
4354 def PEXTRWri : PDIi8<0xC5, MRMSrcReg,
4355 (outs GR32:$dst), (ins VR128:$src1, i32i8imm:$src2),
4356 "pextrw\t{$src2, $src1, $dst|$dst, $src1, $src2}",
4357 [(set GR32:$dst, (X86pextrw (v8i16 VR128:$src1),
4361 let Predicates = [HasAVX] in {
4362 defm VPINSRW : sse2_pinsrw<0>, TB, OpSize, VEX_4V;
4363 def VPINSRWrr64i : Ii8<0xC4, MRMSrcReg, (outs VR128:$dst),
4364 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
4365 "vpinsrw\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
4366 []>, TB, OpSize, VEX_4V;
4369 let Constraints = "$src1 = $dst" in
4370 defm PINSRW : sse2_pinsrw, TB, OpSize, Requires<[HasSSE2]>;
4372 } // ExeDomain = SSEPackedInt
4374 //===---------------------------------------------------------------------===//
4375 // SSE2 - Packed Mask Creation
4376 //===---------------------------------------------------------------------===//
4378 let ExeDomain = SSEPackedInt in {
4380 def VPMOVMSKBrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4381 "pmovmskb\t{$src, $dst|$dst, $src}",
4382 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>, VEX;
4383 def VPMOVMSKBr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR128:$src),
4384 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4386 let Predicates = [HasAVX2] in {
4387 def VPMOVMSKBYrr : VPDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR256:$src),
4388 "pmovmskb\t{$src, $dst|$dst, $src}",
4389 [(set GR32:$dst, (int_x86_avx2_pmovmskb VR256:$src))]>, VEX;
4390 def VPMOVMSKBYr64r : VPDI<0xD7, MRMSrcReg, (outs GR64:$dst), (ins VR256:$src),
4391 "pmovmskb\t{$src, $dst|$dst, $src}", []>, VEX;
4394 def PMOVMSKBrr : PDI<0xD7, MRMSrcReg, (outs GR32:$dst), (ins VR128:$src),
4395 "pmovmskb\t{$src, $dst|$dst, $src}",
4396 [(set GR32:$dst, (int_x86_sse2_pmovmskb_128 VR128:$src))]>;
4398 } // ExeDomain = SSEPackedInt
4400 //===---------------------------------------------------------------------===//
4401 // SSE2 - Conditional Store
4402 //===---------------------------------------------------------------------===//
4404 let ExeDomain = SSEPackedInt in {
4407 def VMASKMOVDQU : VPDI<0xF7, MRMSrcReg, (outs),
4408 (ins VR128:$src, VR128:$mask),
4409 "maskmovdqu\t{$mask, $src|$src, $mask}",
4410 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>, VEX;
4412 def VMASKMOVDQU64 : VPDI<0xF7, MRMSrcReg, (outs),
4413 (ins VR128:$src, VR128:$mask),
4414 "maskmovdqu\t{$mask, $src|$src, $mask}",
4415 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>, VEX;
4418 def MASKMOVDQU : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4419 "maskmovdqu\t{$mask, $src|$src, $mask}",
4420 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, EDI)]>;
4422 def MASKMOVDQU64 : PDI<0xF7, MRMSrcReg, (outs), (ins VR128:$src, VR128:$mask),
4423 "maskmovdqu\t{$mask, $src|$src, $mask}",
4424 [(int_x86_sse2_maskmov_dqu VR128:$src, VR128:$mask, RDI)]>;
4426 } // ExeDomain = SSEPackedInt
4428 //===---------------------------------------------------------------------===//
4429 // SSE2 - Move Doubleword
4430 //===---------------------------------------------------------------------===//
4432 //===---------------------------------------------------------------------===//
4433 // Move Int Doubleword to Packed Double Int
4435 def VMOVDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4436 "movd\t{$src, $dst|$dst, $src}",
4438 (v4i32 (scalar_to_vector GR32:$src)))]>, VEX;
4439 def VMOVDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4440 "movd\t{$src, $dst|$dst, $src}",
4442 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>,
4444 def VMOV64toPQIrr : VRPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4445 "mov{d|q}\t{$src, $dst|$dst, $src}",
4447 (v2i64 (scalar_to_vector GR64:$src)))]>, VEX;
4448 def VMOV64toSDrr : VRPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4449 "mov{d|q}\t{$src, $dst|$dst, $src}",
4450 [(set FR64:$dst, (bitconvert GR64:$src))]>, VEX;
4452 def MOVDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4453 "movd\t{$src, $dst|$dst, $src}",
4455 (v4i32 (scalar_to_vector GR32:$src)))]>;
4456 def MOVDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4457 "movd\t{$src, $dst|$dst, $src}",
4459 (v4i32 (scalar_to_vector (loadi32 addr:$src))))]>;
4460 def MOV64toPQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4461 "mov{d|q}\t{$src, $dst|$dst, $src}",
4463 (v2i64 (scalar_to_vector GR64:$src)))]>;
4464 def MOV64toSDrr : RPDI<0x6E, MRMSrcReg, (outs FR64:$dst), (ins GR64:$src),
4465 "mov{d|q}\t{$src, $dst|$dst, $src}",
4466 [(set FR64:$dst, (bitconvert GR64:$src))]>;
4468 //===---------------------------------------------------------------------===//
4469 // Move Int Doubleword to Single Scalar
4471 def VMOVDI2SSrr : VPDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4472 "movd\t{$src, $dst|$dst, $src}",
4473 [(set FR32:$dst, (bitconvert GR32:$src))]>, VEX;
4475 def VMOVDI2SSrm : VPDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4476 "movd\t{$src, $dst|$dst, $src}",
4477 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>,
4479 def MOVDI2SSrr : PDI<0x6E, MRMSrcReg, (outs FR32:$dst), (ins GR32:$src),
4480 "movd\t{$src, $dst|$dst, $src}",
4481 [(set FR32:$dst, (bitconvert GR32:$src))]>;
4483 def MOVDI2SSrm : PDI<0x6E, MRMSrcMem, (outs FR32:$dst), (ins i32mem:$src),
4484 "movd\t{$src, $dst|$dst, $src}",
4485 [(set FR32:$dst, (bitconvert (loadi32 addr:$src)))]>;
4487 //===---------------------------------------------------------------------===//
4488 // Move Packed Doubleword Int to Packed Double Int
4490 def VMOVPDI2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4491 "movd\t{$src, $dst|$dst, $src}",
4492 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4494 def VMOVPDI2DImr : VPDI<0x7E, MRMDestMem, (outs),
4495 (ins i32mem:$dst, VR128:$src),
4496 "movd\t{$src, $dst|$dst, $src}",
4497 [(store (i32 (vector_extract (v4i32 VR128:$src),
4498 (iPTR 0))), addr:$dst)]>, VEX;
4499 def MOVPDI2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins VR128:$src),
4500 "movd\t{$src, $dst|$dst, $src}",
4501 [(set GR32:$dst, (vector_extract (v4i32 VR128:$src),
4503 def MOVPDI2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, VR128:$src),
4504 "movd\t{$src, $dst|$dst, $src}",
4505 [(store (i32 (vector_extract (v4i32 VR128:$src),
4506 (iPTR 0))), addr:$dst)]>;
4508 //===---------------------------------------------------------------------===//
4509 // Move Packed Doubleword Int first element to Doubleword Int
4511 def VMOVPQIto64rr : I<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4512 "mov{d|q}\t{$src, $dst|$dst, $src}",
4513 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4515 TB, OpSize, VEX, VEX_W, Requires<[HasAVX, In64BitMode]>;
4517 def MOVPQIto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4518 "mov{d|q}\t{$src, $dst|$dst, $src}",
4519 [(set GR64:$dst, (vector_extract (v2i64 VR128:$src),
4522 //===---------------------------------------------------------------------===//
4523 // Bitcast FR64 <-> GR64
4525 let Predicates = [HasAVX] in
4526 def VMOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4527 "vmovq\t{$src, $dst|$dst, $src}",
4528 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>,
4530 def VMOVSDto64rr : VRPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4531 "mov{d|q}\t{$src, $dst|$dst, $src}",
4532 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4533 def VMOVSDto64mr : VRPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4534 "movq\t{$src, $dst|$dst, $src}",
4535 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4537 def MOV64toSDrm : S3SI<0x7E, MRMSrcMem, (outs FR64:$dst), (ins i64mem:$src),
4538 "movq\t{$src, $dst|$dst, $src}",
4539 [(set FR64:$dst, (bitconvert (loadi64 addr:$src)))]>;
4540 def MOVSDto64rr : RPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins FR64:$src),
4541 "mov{d|q}\t{$src, $dst|$dst, $src}",
4542 [(set GR64:$dst, (bitconvert FR64:$src))]>;
4543 def MOVSDto64mr : RPDI<0x7E, MRMDestMem, (outs), (ins i64mem:$dst, FR64:$src),
4544 "movq\t{$src, $dst|$dst, $src}",
4545 [(store (i64 (bitconvert FR64:$src)), addr:$dst)]>;
4547 //===---------------------------------------------------------------------===//
4548 // Move Scalar Single to Double Int
4550 def VMOVSS2DIrr : VPDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4551 "movd\t{$src, $dst|$dst, $src}",
4552 [(set GR32:$dst, (bitconvert FR32:$src))]>, VEX;
4553 def VMOVSS2DImr : VPDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4554 "movd\t{$src, $dst|$dst, $src}",
4555 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>, VEX;
4556 def MOVSS2DIrr : PDI<0x7E, MRMDestReg, (outs GR32:$dst), (ins FR32:$src),
4557 "movd\t{$src, $dst|$dst, $src}",
4558 [(set GR32:$dst, (bitconvert FR32:$src))]>;
4559 def MOVSS2DImr : PDI<0x7E, MRMDestMem, (outs), (ins i32mem:$dst, FR32:$src),
4560 "movd\t{$src, $dst|$dst, $src}",
4561 [(store (i32 (bitconvert FR32:$src)), addr:$dst)]>;
4563 //===---------------------------------------------------------------------===//
4564 // Patterns and instructions to describe movd/movq to XMM register zero-extends
4566 let AddedComplexity = 15 in {
4567 def VMOVZDI2PDIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4568 "movd\t{$src, $dst|$dst, $src}",
4569 [(set VR128:$dst, (v4i32 (X86vzmovl
4570 (v4i32 (scalar_to_vector GR32:$src)))))]>,
4572 def VMOVZQI2PQIrr : VPDI<0x6E, MRMSrcReg, (outs VR128:$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:$src)))))]>,
4578 let AddedComplexity = 15 in {
4579 def MOVZDI2PDIrr : PDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR32:$src),
4580 "movd\t{$src, $dst|$dst, $src}",
4581 [(set VR128:$dst, (v4i32 (X86vzmovl
4582 (v4i32 (scalar_to_vector GR32:$src)))))]>;
4583 def MOVZQI2PQIrr : RPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4584 "mov{d|q}\t{$src, $dst|$dst, $src}", // X86-64 only
4585 [(set VR128:$dst, (v2i64 (X86vzmovl
4586 (v2i64 (scalar_to_vector GR64:$src)))))]>;
4589 let AddedComplexity = 20 in {
4590 def VMOVZDI2PDIrm : VPDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4591 "movd\t{$src, $dst|$dst, $src}",
4593 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4594 (loadi32 addr:$src))))))]>,
4596 def MOVZDI2PDIrm : PDI<0x6E, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
4597 "movd\t{$src, $dst|$dst, $src}",
4599 (v4i32 (X86vzmovl (v4i32 (scalar_to_vector
4600 (loadi32 addr:$src))))))]>;
4603 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4604 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4605 (MOVZDI2PDIrm addr:$src)>;
4606 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4607 (MOVZDI2PDIrm addr:$src)>;
4608 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4609 (MOVZDI2PDIrm addr:$src)>;
4612 let Predicates = [HasAVX] in {
4613 // AVX 128-bit movd/movq instruction write zeros in the high 128-bit part.
4614 let AddedComplexity = 20 in {
4615 def : Pat<(v4i32 (X86vzmovl (loadv4i32 addr:$src))),
4616 (VMOVZDI2PDIrm addr:$src)>;
4617 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv4f32 addr:$src)))),
4618 (VMOVZDI2PDIrm addr:$src)>;
4619 def : Pat<(v4i32 (X86vzmovl (bc_v4i32 (loadv2i64 addr:$src)))),
4620 (VMOVZDI2PDIrm addr:$src)>;
4622 // Use regular 128-bit instructions to match 256-bit scalar_to_vec+zext.
4623 def : Pat<(v8i32 (X86vzmovl (insert_subvector undef,
4624 (v4i32 (scalar_to_vector GR32:$src)),(i32 0)))),
4625 (SUBREG_TO_REG (i32 0), (VMOVZDI2PDIrr GR32:$src), sub_xmm)>;
4626 def : Pat<(v4i64 (X86vzmovl (insert_subvector undef,
4627 (v2i64 (scalar_to_vector GR64:$src)),(i32 0)))),
4628 (SUBREG_TO_REG (i64 0), (VMOVZQI2PQIrr GR64:$src), sub_xmm)>;
4631 // These are the correct encodings of the instructions so that we know how to
4632 // read correct assembly, even though we continue to emit the wrong ones for
4633 // compatibility with Darwin's buggy assembler.
4634 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4635 (MOV64toPQIrr VR128:$dst, GR64:$src), 0>;
4636 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4637 (MOV64toSDrr FR64:$dst, GR64:$src), 0>;
4638 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4639 (MOVPQIto64rr GR64:$dst, VR128:$src), 0>;
4640 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4641 (MOVSDto64rr GR64:$dst, FR64:$src), 0>;
4642 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4643 (VMOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4644 def : InstAlias<"movq\t{$src, $dst|$dst, $src}",
4645 (MOVZQI2PQIrr VR128:$dst, GR64:$src), 0>;
4647 //===---------------------------------------------------------------------===//
4648 // SSE2 - Move Quadword
4649 //===---------------------------------------------------------------------===//
4651 //===---------------------------------------------------------------------===//
4652 // Move Quadword Int to Packed Quadword Int
4654 def VMOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4655 "vmovq\t{$src, $dst|$dst, $src}",
4657 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4658 VEX, Requires<[HasAVX]>;
4659 def MOVQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4660 "movq\t{$src, $dst|$dst, $src}",
4662 (v2i64 (scalar_to_vector (loadi64 addr:$src))))]>, XS,
4663 Requires<[HasSSE2]>; // SSE2 instruction with XS Prefix
4665 //===---------------------------------------------------------------------===//
4666 // Move Packed Quadword Int to Quadword Int
4668 def VMOVPQI2QImr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4669 "movq\t{$src, $dst|$dst, $src}",
4670 [(store (i64 (vector_extract (v2i64 VR128:$src),
4671 (iPTR 0))), addr:$dst)]>, VEX;
4672 def MOVPQI2QImr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4673 "movq\t{$src, $dst|$dst, $src}",
4674 [(store (i64 (vector_extract (v2i64 VR128:$src),
4675 (iPTR 0))), addr:$dst)]>;
4677 //===---------------------------------------------------------------------===//
4678 // Store / copy lower 64-bits of a XMM register.
4680 def VMOVLQ128mr : VPDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4681 "movq\t{$src, $dst|$dst, $src}",
4682 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>, VEX;
4683 def MOVLQ128mr : PDI<0xD6, MRMDestMem, (outs), (ins i64mem:$dst, VR128:$src),
4684 "movq\t{$src, $dst|$dst, $src}",
4685 [(int_x86_sse2_storel_dq addr:$dst, VR128:$src)]>;
4687 let AddedComplexity = 20 in
4688 def VMOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4689 "vmovq\t{$src, $dst|$dst, $src}",
4691 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4692 (loadi64 addr:$src))))))]>,
4693 XS, VEX, Requires<[HasAVX]>;
4695 let AddedComplexity = 20 in
4696 def MOVZQI2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
4697 "movq\t{$src, $dst|$dst, $src}",
4699 (v2i64 (X86vzmovl (v2i64 (scalar_to_vector
4700 (loadi64 addr:$src))))))]>,
4701 XS, Requires<[HasSSE2]>;
4703 let Predicates = [HasSSE2], AddedComplexity = 20 in {
4704 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4705 (MOVZQI2PQIrm addr:$src)>;
4706 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4707 (MOVZQI2PQIrm addr:$src)>;
4708 def : Pat<(v2i64 (X86vzload addr:$src)), (MOVZQI2PQIrm addr:$src)>;
4711 let Predicates = [HasAVX], AddedComplexity = 20 in {
4712 def : Pat<(v2i64 (X86vzmovl (loadv2i64 addr:$src))),
4713 (VMOVZQI2PQIrm addr:$src)>;
4714 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4f32 addr:$src)))),
4715 (VMOVZQI2PQIrm addr:$src)>;
4716 def : Pat<(v2i64 (X86vzload addr:$src)),
4717 (VMOVZQI2PQIrm addr:$src)>;
4720 //===---------------------------------------------------------------------===//
4721 // Moving from XMM to XMM and clear upper 64 bits. Note, there is a bug in
4722 // IA32 document. movq xmm1, xmm2 does clear the high bits.
4724 let AddedComplexity = 15 in
4725 def VMOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4726 "vmovq\t{$src, $dst|$dst, $src}",
4727 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4728 XS, VEX, Requires<[HasAVX]>;
4729 let AddedComplexity = 15 in
4730 def MOVZPQILo2PQIrr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4731 "movq\t{$src, $dst|$dst, $src}",
4732 [(set VR128:$dst, (v2i64 (X86vzmovl (v2i64 VR128:$src))))]>,
4733 XS, Requires<[HasSSE2]>;
4735 let AddedComplexity = 20 in
4736 def VMOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4737 "vmovq\t{$src, $dst|$dst, $src}",
4738 [(set VR128:$dst, (v2i64 (X86vzmovl
4739 (loadv2i64 addr:$src))))]>,
4740 XS, VEX, Requires<[HasAVX]>;
4741 let AddedComplexity = 20 in {
4742 def MOVZPQILo2PQIrm : I<0x7E, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
4743 "movq\t{$src, $dst|$dst, $src}",
4744 [(set VR128:$dst, (v2i64 (X86vzmovl
4745 (loadv2i64 addr:$src))))]>,
4746 XS, Requires<[HasSSE2]>;
4749 let AddedComplexity = 20 in {
4750 let Predicates = [HasSSE2] in {
4751 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4752 (MOVZPQILo2PQIrm addr:$src)>;
4753 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4754 (MOVZPQILo2PQIrr VR128:$src)>;
4756 let Predicates = [HasAVX] in {
4757 def : Pat<(v2i64 (X86vzmovl (bc_v2i64 (loadv4i32 addr:$src)))),
4758 (VMOVZPQILo2PQIrm addr:$src)>;
4759 def : Pat<(v2f64 (X86vzmovl (v2f64 VR128:$src))),
4760 (VMOVZPQILo2PQIrr VR128:$src)>;
4764 // Instructions to match in the assembler
4765 def VMOVQs64rr : VPDI<0x6E, MRMSrcReg, (outs VR128:$dst), (ins GR64:$src),
4766 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4767 def VMOVQd64rr : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4768 "movq\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4769 // Recognize "movd" with GR64 destination, but encode as a "movq"
4770 def VMOVQd64rr_alt : VPDI<0x7E, MRMDestReg, (outs GR64:$dst), (ins VR128:$src),
4771 "movd\t{$src, $dst|$dst, $src}", []>, VEX, VEX_W;
4773 // Instructions for the disassembler
4774 // xr = XMM register
4777 let Predicates = [HasAVX] in
4778 def VMOVQxrxr: I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4779 "vmovq\t{$src, $dst|$dst, $src}", []>, VEX, XS;
4780 def MOVQxrxr : I<0x7E, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4781 "movq\t{$src, $dst|$dst, $src}", []>, XS;
4783 //===---------------------------------------------------------------------===//
4784 // SSE3 - Conversion Instructions
4785 //===---------------------------------------------------------------------===//
4787 // Convert Packed Double FP to Packed DW Integers
4788 let Predicates = [HasAVX] in {
4789 // The assembler can recognize rr 256-bit instructions by seeing a ymm
4790 // register, but the same isn't true when using memory operands instead.
4791 // Provide other assembly rr and rm forms to address this explicitly.
4792 def VCVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4793 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4794 def VCVTPD2DQXrYr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4795 "vcvtpd2dq\t{$src, $dst|$dst, $src}", []>, VEX;
4798 def VCVTPD2DQXrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4799 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4800 def VCVTPD2DQXrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4801 "vcvtpd2dqx\t{$src, $dst|$dst, $src}", []>, VEX;
4804 def VCVTPD2DQYrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR256:$src),
4805 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX;
4806 def VCVTPD2DQYrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f256mem:$src),
4807 "vcvtpd2dqy\t{$src, $dst|$dst, $src}", []>, VEX, VEX_L;
4810 def CVTPD2DQrm : S3DI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4811 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4812 def CVTPD2DQrr : S3DI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4813 "cvtpd2dq\t{$src, $dst|$dst, $src}", []>;
4815 def : Pat<(v4i32 (fp_to_sint (v4f64 VR256:$src))),
4816 (VCVTPD2DQYrr VR256:$src)>;
4817 def : Pat<(v4i32 (fp_to_sint (memopv4f64 addr:$src))),
4818 (VCVTPD2DQYrm addr:$src)>;
4820 // Convert Packed DW Integers to Packed Double FP
4821 let Predicates = [HasAVX] in {
4822 def VCVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4823 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4824 def VCVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4825 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4826 def VCVTDQ2PDYrm : S3SI<0xE6, MRMSrcMem, (outs VR256:$dst), (ins f128mem:$src),
4827 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4828 def VCVTDQ2PDYrr : S3SI<0xE6, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
4829 "vcvtdq2pd\t{$src, $dst|$dst, $src}", []>, VEX;
4832 def CVTDQ2PDrm : S3SI<0xE6, MRMSrcMem, (outs VR128:$dst), (ins f128mem:$src),
4833 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4834 def CVTDQ2PDrr : S3SI<0xE6, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4835 "cvtdq2pd\t{$src, $dst|$dst, $src}", []>;
4837 // AVX 256-bit register conversion intrinsics
4838 def : Pat<(int_x86_avx_cvtdq2_pd_256 VR128:$src),
4839 (VCVTDQ2PDYrr VR128:$src)>;
4840 def : Pat<(int_x86_avx_cvtdq2_pd_256 (bitconvert (memopv2i64 addr:$src))),
4841 (VCVTDQ2PDYrm addr:$src)>;
4843 def : Pat<(int_x86_avx_cvt_pd2dq_256 VR256:$src),
4844 (VCVTPD2DQYrr VR256:$src)>;
4845 def : Pat<(int_x86_avx_cvt_pd2dq_256 (memopv4f64 addr:$src)),
4846 (VCVTPD2DQYrm addr:$src)>;
4848 def : Pat<(v4f64 (sint_to_fp (v4i32 VR128:$src))),
4849 (VCVTDQ2PDYrr VR128:$src)>;
4850 def : Pat<(v4f64 (sint_to_fp (bc_v4i32 (memopv2i64 addr:$src)))),
4851 (VCVTDQ2PDYrm addr:$src)>;
4853 //===---------------------------------------------------------------------===//
4854 // SSE3 - Replicate Single FP - MOVSHDUP and MOVSLDUP
4855 //===---------------------------------------------------------------------===//
4856 multiclass sse3_replicate_sfp<bits<8> op, SDNode OpNode, string OpcodeStr,
4857 ValueType vt, RegisterClass RC, PatFrag mem_frag,
4858 X86MemOperand x86memop> {
4859 def rr : S3SI<op, MRMSrcReg, (outs RC:$dst), (ins RC:$src),
4860 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4861 [(set RC:$dst, (vt (OpNode RC:$src)))]>;
4862 def rm : S3SI<op, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
4863 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4864 [(set RC:$dst, (OpNode (mem_frag addr:$src)))]>;
4867 let Predicates = [HasAVX] in {
4868 defm VMOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4869 v4f32, VR128, memopv4f32, f128mem>, VEX;
4870 defm VMOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4871 v4f32, VR128, memopv4f32, f128mem>, VEX;
4872 defm VMOVSHDUPY : sse3_replicate_sfp<0x16, X86Movshdup, "vmovshdup",
4873 v8f32, VR256, memopv8f32, f256mem>, VEX;
4874 defm VMOVSLDUPY : sse3_replicate_sfp<0x12, X86Movsldup, "vmovsldup",
4875 v8f32, VR256, memopv8f32, f256mem>, VEX;
4877 defm MOVSHDUP : sse3_replicate_sfp<0x16, X86Movshdup, "movshdup", v4f32, VR128,
4878 memopv4f32, f128mem>;
4879 defm MOVSLDUP : sse3_replicate_sfp<0x12, X86Movsldup, "movsldup", v4f32, VR128,
4880 memopv4f32, f128mem>;
4882 let Predicates = [HasSSE3] in {
4883 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4884 (MOVSHDUPrr VR128:$src)>;
4885 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4886 (MOVSHDUPrm addr:$src)>;
4887 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4888 (MOVSLDUPrr VR128:$src)>;
4889 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4890 (MOVSLDUPrm addr:$src)>;
4893 let Predicates = [HasAVX] in {
4894 def : Pat<(v4i32 (X86Movshdup VR128:$src)),
4895 (VMOVSHDUPrr VR128:$src)>;
4896 def : Pat<(v4i32 (X86Movshdup (bc_v4i32 (memopv2i64 addr:$src)))),
4897 (VMOVSHDUPrm addr:$src)>;
4898 def : Pat<(v4i32 (X86Movsldup VR128:$src)),
4899 (VMOVSLDUPrr VR128:$src)>;
4900 def : Pat<(v4i32 (X86Movsldup (bc_v4i32 (memopv2i64 addr:$src)))),
4901 (VMOVSLDUPrm addr:$src)>;
4902 def : Pat<(v8i32 (X86Movshdup VR256:$src)),
4903 (VMOVSHDUPYrr VR256:$src)>;
4904 def : Pat<(v8i32 (X86Movshdup (bc_v8i32 (memopv4i64 addr:$src)))),
4905 (VMOVSHDUPYrm addr:$src)>;
4906 def : Pat<(v8i32 (X86Movsldup VR256:$src)),
4907 (VMOVSLDUPYrr VR256:$src)>;
4908 def : Pat<(v8i32 (X86Movsldup (bc_v8i32 (memopv4i64 addr:$src)))),
4909 (VMOVSLDUPYrm addr:$src)>;
4912 //===---------------------------------------------------------------------===//
4913 // SSE3 - Replicate Double FP - MOVDDUP
4914 //===---------------------------------------------------------------------===//
4916 multiclass sse3_replicate_dfp<string OpcodeStr> {
4917 def rr : S3DI<0x12, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
4918 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4919 [(set VR128:$dst,(v2f64 (movddup VR128:$src, (undef))))]>;
4920 def rm : S3DI<0x12, MRMSrcMem, (outs VR128:$dst), (ins f64mem:$src),
4921 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4923 (v2f64 (movddup (scalar_to_vector (loadf64 addr:$src)),
4927 // FIXME: Merge with above classe when there're patterns for the ymm version
4928 multiclass sse3_replicate_dfp_y<string OpcodeStr> {
4929 let Predicates = [HasAVX] in {
4930 def rr : S3DI<0x12, MRMSrcReg, (outs VR256:$dst), (ins VR256:$src),
4931 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4933 def rm : S3DI<0x12, MRMSrcMem, (outs VR256:$dst), (ins f256mem:$src),
4934 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
4939 defm MOVDDUP : sse3_replicate_dfp<"movddup">;
4940 defm VMOVDDUP : sse3_replicate_dfp<"vmovddup">, VEX;
4941 defm VMOVDDUPY : sse3_replicate_dfp_y<"vmovddup">, VEX;
4943 let Predicates = [HasSSE3] in {
4944 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4946 (MOVDDUPrm addr:$src)>;
4947 let AddedComplexity = 5 in {
4948 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4949 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4950 (MOVDDUPrm addr:$src)>;
4951 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (MOVDDUPrm addr:$src)>;
4952 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4953 (MOVDDUPrm addr:$src)>;
4955 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4956 (MOVDDUPrm addr:$src)>;
4957 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4958 (MOVDDUPrm addr:$src)>;
4959 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4960 (MOVDDUPrm addr:$src)>;
4961 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4962 (MOVDDUPrm addr:$src)>;
4963 def : Pat<(X86Movddup (bc_v2f64
4964 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4965 (MOVDDUPrm addr:$src)>;
4968 let Predicates = [HasAVX] in {
4969 def : Pat<(movddup (bc_v2f64 (v2i64 (scalar_to_vector (loadi64 addr:$src)))),
4971 (VMOVDDUPrm addr:$src)>;
4972 let AddedComplexity = 5 in {
4973 def : Pat<(movddup (memopv2f64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4974 def : Pat<(movddup (bc_v4f32 (memopv2f64 addr:$src)), (undef)),
4975 (VMOVDDUPrm addr:$src)>;
4976 def : Pat<(movddup (memopv2i64 addr:$src), (undef)), (VMOVDDUPrm addr:$src)>;
4977 def : Pat<(movddup (bc_v4i32 (memopv2i64 addr:$src)), (undef)),
4978 (VMOVDDUPrm addr:$src)>;
4980 def : Pat<(X86Movddup (memopv2f64 addr:$src)),
4981 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4982 def : Pat<(X86Movddup (bc_v2f64 (memopv4f32 addr:$src))),
4983 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4984 def : Pat<(X86Movddup (bc_v2f64 (memopv2i64 addr:$src))),
4985 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4986 def : Pat<(X86Movddup (v2f64 (scalar_to_vector (loadf64 addr:$src)))),
4987 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4988 def : Pat<(X86Movddup (bc_v2f64
4989 (v2i64 (scalar_to_vector (loadi64 addr:$src))))),
4990 (VMOVDDUPrm addr:$src)>, Requires<[HasAVX]>;
4993 def : Pat<(X86Movddup (memopv4f64 addr:$src)),
4994 (VMOVDDUPYrm addr:$src)>;
4995 def : Pat<(X86Movddup (memopv4i64 addr:$src)),
4996 (VMOVDDUPYrm addr:$src)>;
4997 def : Pat<(X86Movddup (v4f64 (scalar_to_vector (loadf64 addr:$src)))),
4998 (VMOVDDUPYrm addr:$src)>;
4999 def : Pat<(X86Movddup (v4i64 (scalar_to_vector (loadi64 addr:$src)))),
5000 (VMOVDDUPYrm addr:$src)>;
5001 def : Pat<(X86Movddup (v4f64 VR256:$src)),
5002 (VMOVDDUPYrr VR256:$src)>;
5003 def : Pat<(X86Movddup (v4i64 VR256:$src)),
5004 (VMOVDDUPYrr VR256:$src)>;
5007 //===---------------------------------------------------------------------===//
5008 // SSE3 - Move Unaligned Integer
5009 //===---------------------------------------------------------------------===//
5011 let Predicates = [HasAVX] in {
5012 def VLDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5013 "vlddqu\t{$src, $dst|$dst, $src}",
5014 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>, VEX;
5015 def VLDDQUYrm : S3DI<0xF0, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
5016 "vlddqu\t{$src, $dst|$dst, $src}",
5017 [(set VR256:$dst, (int_x86_avx_ldu_dq_256 addr:$src))]>, VEX;
5019 def LDDQUrm : S3DI<0xF0, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
5020 "lddqu\t{$src, $dst|$dst, $src}",
5021 [(set VR128:$dst, (int_x86_sse3_ldu_dq addr:$src))]>;
5023 //===---------------------------------------------------------------------===//
5024 // SSE3 - Arithmetic
5025 //===---------------------------------------------------------------------===//
5027 multiclass sse3_addsub<Intrinsic Int, string OpcodeStr, RegisterClass RC,
5028 X86MemOperand x86memop, bit Is2Addr = 1> {
5029 def rr : I<0xD0, MRMSrcReg,
5030 (outs RC:$dst), (ins RC:$src1, RC:$src2),
5032 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5033 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5034 [(set RC:$dst, (Int RC:$src1, RC:$src2))]>;
5035 def rm : I<0xD0, MRMSrcMem,
5036 (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5038 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5039 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5040 [(set RC:$dst, (Int RC:$src1, (memop addr:$src2)))]>;
5043 let Predicates = [HasAVX] in {
5044 let ExeDomain = SSEPackedSingle in {
5045 defm VADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "vaddsubps", VR128,
5046 f128mem, 0>, TB, XD, VEX_4V;
5047 defm VADDSUBPSY : sse3_addsub<int_x86_avx_addsub_ps_256, "vaddsubps", VR256,
5048 f256mem, 0>, TB, XD, VEX_4V;
5050 let ExeDomain = SSEPackedDouble in {
5051 defm VADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "vaddsubpd", VR128,
5052 f128mem, 0>, TB, OpSize, VEX_4V;
5053 defm VADDSUBPDY : sse3_addsub<int_x86_avx_addsub_pd_256, "vaddsubpd", VR256,
5054 f256mem, 0>, TB, OpSize, VEX_4V;
5057 let Constraints = "$src1 = $dst", Predicates = [HasSSE3] in {
5058 let ExeDomain = SSEPackedSingle in
5059 defm ADDSUBPS : sse3_addsub<int_x86_sse3_addsub_ps, "addsubps", VR128,
5061 let ExeDomain = SSEPackedDouble in
5062 defm ADDSUBPD : sse3_addsub<int_x86_sse3_addsub_pd, "addsubpd", VR128,
5063 f128mem>, TB, OpSize;
5066 //===---------------------------------------------------------------------===//
5067 // SSE3 Instructions
5068 //===---------------------------------------------------------------------===//
5071 multiclass S3D_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5072 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5073 def rr : S3DI<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5075 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5076 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5077 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5079 def rm : S3DI<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5081 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5082 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5083 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5085 multiclass S3_Int<bits<8> o, string OpcodeStr, ValueType vt, RegisterClass RC,
5086 X86MemOperand x86memop, SDNode OpNode, bit Is2Addr = 1> {
5087 def rr : S3I<o, MRMSrcReg, (outs RC:$dst), (ins RC:$src1, RC:$src2),
5089 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5090 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5091 [(set RC:$dst, (vt (OpNode RC:$src1, RC:$src2)))]>;
5093 def rm : S3I<o, MRMSrcMem, (outs RC:$dst), (ins RC:$src1, x86memop:$src2),
5095 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5096 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5097 [(set RC:$dst, (vt (OpNode RC:$src1, (memop addr:$src2))))]>;
5100 let Predicates = [HasAVX] in {
5101 let ExeDomain = SSEPackedSingle in {
5102 defm VHADDPS : S3D_Int<0x7C, "vhaddps", v4f32, VR128, f128mem,
5103 X86fhadd, 0>, VEX_4V;
5104 defm VHSUBPS : S3D_Int<0x7D, "vhsubps", v4f32, VR128, f128mem,
5105 X86fhsub, 0>, VEX_4V;
5106 defm VHADDPSY : S3D_Int<0x7C, "vhaddps", v8f32, VR256, f256mem,
5107 X86fhadd, 0>, VEX_4V;
5108 defm VHSUBPSY : S3D_Int<0x7D, "vhsubps", v8f32, VR256, f256mem,
5109 X86fhsub, 0>, VEX_4V;
5111 let ExeDomain = SSEPackedDouble in {
5112 defm VHADDPD : S3_Int <0x7C, "vhaddpd", v2f64, VR128, f128mem,
5113 X86fhadd, 0>, VEX_4V;
5114 defm VHSUBPD : S3_Int <0x7D, "vhsubpd", v2f64, VR128, f128mem,
5115 X86fhsub, 0>, VEX_4V;
5116 defm VHADDPDY : S3_Int <0x7C, "vhaddpd", v4f64, VR256, f256mem,
5117 X86fhadd, 0>, VEX_4V;
5118 defm VHSUBPDY : S3_Int <0x7D, "vhsubpd", v4f64, VR256, f256mem,
5119 X86fhsub, 0>, VEX_4V;
5123 let Constraints = "$src1 = $dst" in {
5124 let ExeDomain = SSEPackedSingle in {
5125 defm HADDPS : S3D_Int<0x7C, "haddps", v4f32, VR128, f128mem, X86fhadd>;
5126 defm HSUBPS : S3D_Int<0x7D, "hsubps", v4f32, VR128, f128mem, X86fhsub>;
5128 let ExeDomain = SSEPackedDouble in {
5129 defm HADDPD : S3_Int<0x7C, "haddpd", v2f64, VR128, f128mem, X86fhadd>;
5130 defm HSUBPD : S3_Int<0x7D, "hsubpd", v2f64, VR128, f128mem, X86fhsub>;
5134 //===---------------------------------------------------------------------===//
5135 // SSSE3 - Packed Absolute Instructions
5136 //===---------------------------------------------------------------------===//
5139 /// SS3I_unop_rm_int - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5140 multiclass SS3I_unop_rm_int<bits<8> opc, string OpcodeStr,
5141 Intrinsic IntId128> {
5142 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5144 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5145 [(set VR128:$dst, (IntId128 VR128:$src))]>,
5148 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5150 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5153 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
5156 /// SS3I_unop_rm_int_y - Simple SSSE3 unary op whose type can be v*{i8,i16,i32}.
5157 multiclass SS3I_unop_rm_int_y<bits<8> opc, string OpcodeStr,
5158 Intrinsic IntId256> {
5159 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5161 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5162 [(set VR256:$dst, (IntId256 VR256:$src))]>,
5165 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5167 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5170 (bitconvert (memopv4i64 addr:$src))))]>, OpSize;
5173 let Predicates = [HasAVX] in {
5174 defm VPABSB : SS3I_unop_rm_int<0x1C, "vpabsb",
5175 int_x86_ssse3_pabs_b_128>, VEX;
5176 defm VPABSW : SS3I_unop_rm_int<0x1D, "vpabsw",
5177 int_x86_ssse3_pabs_w_128>, VEX;
5178 defm VPABSD : SS3I_unop_rm_int<0x1E, "vpabsd",
5179 int_x86_ssse3_pabs_d_128>, VEX;
5182 let Predicates = [HasAVX2] in {
5183 defm VPABSB : SS3I_unop_rm_int_y<0x1C, "vpabsb",
5184 int_x86_avx2_pabs_b>, VEX;
5185 defm VPABSW : SS3I_unop_rm_int_y<0x1D, "vpabsw",
5186 int_x86_avx2_pabs_w>, VEX;
5187 defm VPABSD : SS3I_unop_rm_int_y<0x1E, "vpabsd",
5188 int_x86_avx2_pabs_d>, VEX;
5191 defm PABSB : SS3I_unop_rm_int<0x1C, "pabsb",
5192 int_x86_ssse3_pabs_b_128>;
5193 defm PABSW : SS3I_unop_rm_int<0x1D, "pabsw",
5194 int_x86_ssse3_pabs_w_128>;
5195 defm PABSD : SS3I_unop_rm_int<0x1E, "pabsd",
5196 int_x86_ssse3_pabs_d_128>;
5198 //===---------------------------------------------------------------------===//
5199 // SSSE3 - Packed Binary Operator Instructions
5200 //===---------------------------------------------------------------------===//
5202 /// SS3I_binop_rm_int - Simple SSSE3 bin op whose type can be v*{i8,i16,i32}.
5203 multiclass SS3I_binop_rm_int<bits<8> opc, string OpcodeStr,
5204 Intrinsic IntId128, bit Is2Addr = 1> {
5205 let isCommutable = 1 in
5206 def rr128 : SS38I<opc, MRMSrcReg, (outs VR128:$dst),
5207 (ins VR128:$src1, VR128:$src2),
5209 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5210 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5211 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
5213 def rm128 : SS38I<opc, MRMSrcMem, (outs VR128:$dst),
5214 (ins VR128:$src1, i128mem:$src2),
5216 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
5217 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
5219 (IntId128 VR128:$src1,
5220 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
5223 multiclass SS3I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
5224 Intrinsic IntId256> {
5225 let isCommutable = 1 in
5226 def rr256 : SS38I<opc, MRMSrcReg, (outs VR256:$dst),
5227 (ins VR256:$src1, VR256:$src2),
5228 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5229 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
5231 def rm256 : SS38I<opc, MRMSrcMem, (outs VR256:$dst),
5232 (ins VR256:$src1, i256mem:$src2),
5233 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5235 (IntId256 VR256:$src1,
5236 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
5239 let ImmT = NoImm, Predicates = [HasAVX] in {
5240 let isCommutable = 0 in {
5241 defm VPHADDW : SS3I_binop_rm_int<0x01, "vphaddw",
5242 int_x86_ssse3_phadd_w_128, 0>, VEX_4V;
5243 defm VPHADDD : SS3I_binop_rm_int<0x02, "vphaddd",
5244 int_x86_ssse3_phadd_d_128, 0>, VEX_4V;
5245 defm VPHADDSW : SS3I_binop_rm_int<0x03, "vphaddsw",
5246 int_x86_ssse3_phadd_sw_128, 0>, VEX_4V;
5247 defm VPHSUBW : SS3I_binop_rm_int<0x05, "vphsubw",
5248 int_x86_ssse3_phsub_w_128, 0>, VEX_4V;
5249 defm VPHSUBD : SS3I_binop_rm_int<0x06, "vphsubd",
5250 int_x86_ssse3_phsub_d_128, 0>, VEX_4V;
5251 defm VPHSUBSW : SS3I_binop_rm_int<0x07, "vphsubsw",
5252 int_x86_ssse3_phsub_sw_128, 0>, VEX_4V;
5253 defm VPMADDUBSW : SS3I_binop_rm_int<0x04, "vpmaddubsw",
5254 int_x86_ssse3_pmadd_ub_sw_128, 0>, VEX_4V;
5255 defm VPSHUFB : SS3I_binop_rm_int<0x00, "vpshufb",
5256 int_x86_ssse3_pshuf_b_128, 0>, VEX_4V;
5257 defm VPSIGNB : SS3I_binop_rm_int<0x08, "vpsignb",
5258 int_x86_ssse3_psign_b_128, 0>, VEX_4V;
5259 defm VPSIGNW : SS3I_binop_rm_int<0x09, "vpsignw",
5260 int_x86_ssse3_psign_w_128, 0>, VEX_4V;
5261 defm VPSIGND : SS3I_binop_rm_int<0x0A, "vpsignd",
5262 int_x86_ssse3_psign_d_128, 0>, VEX_4V;
5264 defm VPMULHRSW : SS3I_binop_rm_int<0x0B, "vpmulhrsw",
5265 int_x86_ssse3_pmul_hr_sw_128, 0>, VEX_4V;
5268 let ImmT = NoImm, Predicates = [HasAVX2] in {
5269 let isCommutable = 0 in {
5270 defm VPHADDW : SS3I_binop_rm_int_y<0x01, "vphaddw",
5271 int_x86_avx2_phadd_w>, VEX_4V;
5272 defm VPHADDD : SS3I_binop_rm_int_y<0x02, "vphaddd",
5273 int_x86_avx2_phadd_d>, VEX_4V;
5274 defm VPHADDSW : SS3I_binop_rm_int_y<0x03, "vphaddsw",
5275 int_x86_avx2_phadd_sw>, VEX_4V;
5276 defm VPHSUBW : SS3I_binop_rm_int_y<0x05, "vphsubw",
5277 int_x86_avx2_phsub_w>, VEX_4V;
5278 defm VPHSUBD : SS3I_binop_rm_int_y<0x06, "vphsubd",
5279 int_x86_avx2_phsub_d>, VEX_4V;
5280 defm VPHSUBSW : SS3I_binop_rm_int_y<0x07, "vphsubsw",
5281 int_x86_avx2_phsub_sw>, VEX_4V;
5282 defm VPMADDUBSW : SS3I_binop_rm_int_y<0x04, "vpmaddubsw",
5283 int_x86_avx2_pmadd_ub_sw>, VEX_4V;
5284 defm VPSHUFB : SS3I_binop_rm_int_y<0x00, "vpshufb",
5285 int_x86_avx2_pshuf_b>, VEX_4V;
5286 defm VPSIGNB : SS3I_binop_rm_int_y<0x08, "vpsignb",
5287 int_x86_avx2_psign_b>, VEX_4V;
5288 defm VPSIGNW : SS3I_binop_rm_int_y<0x09, "vpsignw",
5289 int_x86_avx2_psign_w>, VEX_4V;
5290 defm VPSIGND : SS3I_binop_rm_int_y<0x0A, "vpsignd",
5291 int_x86_avx2_psign_d>, VEX_4V;
5293 defm VPMULHRSW : SS3I_binop_rm_int_y<0x0B, "vpmulhrsw",
5294 int_x86_avx2_pmul_hr_sw>, VEX_4V;
5297 // None of these have i8 immediate fields.
5298 let ImmT = NoImm, Constraints = "$src1 = $dst" in {
5299 let isCommutable = 0 in {
5300 defm PHADDW : SS3I_binop_rm_int<0x01, "phaddw",
5301 int_x86_ssse3_phadd_w_128>;
5302 defm PHADDD : SS3I_binop_rm_int<0x02, "phaddd",
5303 int_x86_ssse3_phadd_d_128>;
5304 defm PHADDSW : SS3I_binop_rm_int<0x03, "phaddsw",
5305 int_x86_ssse3_phadd_sw_128>;
5306 defm PHSUBW : SS3I_binop_rm_int<0x05, "phsubw",
5307 int_x86_ssse3_phsub_w_128>;
5308 defm PHSUBD : SS3I_binop_rm_int<0x06, "phsubd",
5309 int_x86_ssse3_phsub_d_128>;
5310 defm PHSUBSW : SS3I_binop_rm_int<0x07, "phsubsw",
5311 int_x86_ssse3_phsub_sw_128>;
5312 defm PMADDUBSW : SS3I_binop_rm_int<0x04, "pmaddubsw",
5313 int_x86_ssse3_pmadd_ub_sw_128>;
5314 defm PSHUFB : SS3I_binop_rm_int<0x00, "pshufb",
5315 int_x86_ssse3_pshuf_b_128>;
5316 defm PSIGNB : SS3I_binop_rm_int<0x08, "psignb",
5317 int_x86_ssse3_psign_b_128>;
5318 defm PSIGNW : SS3I_binop_rm_int<0x09, "psignw",
5319 int_x86_ssse3_psign_w_128>;
5320 defm PSIGND : SS3I_binop_rm_int<0x0A, "psignd",
5321 int_x86_ssse3_psign_d_128>;
5323 defm PMULHRSW : SS3I_binop_rm_int<0x0B, "pmulhrsw",
5324 int_x86_ssse3_pmul_hr_sw_128>;
5327 let Predicates = [HasSSSE3] in {
5328 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5329 (PSHUFBrr128 VR128:$src, VR128:$mask)>;
5330 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5331 (PSHUFBrm128 VR128:$src, addr:$mask)>;
5333 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5334 (PSIGNBrr128 VR128:$src1, VR128:$src2)>;
5335 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5336 (PSIGNWrr128 VR128:$src1, VR128:$src2)>;
5337 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5338 (PSIGNDrr128 VR128:$src1, VR128:$src2)>;
5340 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5341 (PHADDWrr128 VR128:$src1, VR128:$src2)>;
5342 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5343 (PHADDDrr128 VR128:$src1, VR128:$src2)>;
5344 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5345 (PHSUBWrr128 VR128:$src1, VR128:$src2)>;
5346 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5347 (PHSUBDrr128 VR128:$src1, VR128:$src2)>;
5350 let Predicates = [HasAVX] in {
5351 def : Pat<(X86pshufb VR128:$src, VR128:$mask),
5352 (VPSHUFBrr128 VR128:$src, VR128:$mask)>;
5353 def : Pat<(X86pshufb VR128:$src, (bc_v16i8 (memopv2i64 addr:$mask))),
5354 (VPSHUFBrm128 VR128:$src, addr:$mask)>;
5356 def : Pat<(v16i8 (X86psign VR128:$src1, VR128:$src2)),
5357 (VPSIGNBrr128 VR128:$src1, VR128:$src2)>;
5358 def : Pat<(v8i16 (X86psign VR128:$src1, VR128:$src2)),
5359 (VPSIGNWrr128 VR128:$src1, VR128:$src2)>;
5360 def : Pat<(v4i32 (X86psign VR128:$src1, VR128:$src2)),
5361 (VPSIGNDrr128 VR128:$src1, VR128:$src2)>;
5363 def : Pat<(v8i16 (X86hadd VR128:$src1, VR128:$src2)),
5364 (VPHADDWrr128 VR128:$src1, VR128:$src2)>;
5365 def : Pat<(v4i32 (X86hadd VR128:$src1, VR128:$src2)),
5366 (VPHADDDrr128 VR128:$src1, VR128:$src2)>;
5367 def : Pat<(v8i16 (X86hsub VR128:$src1, VR128:$src2)),
5368 (VPHSUBWrr128 VR128:$src1, VR128:$src2)>;
5369 def : Pat<(v4i32 (X86hsub VR128:$src1, VR128:$src2)),
5370 (VPHSUBDrr128 VR128:$src1, VR128:$src2)>;
5373 let Predicates = [HasAVX2] in {
5374 def : Pat<(v32i8 (X86psign VR256:$src1, VR256:$src2)),
5375 (VPSIGNBrr256 VR256:$src1, VR256:$src2)>;
5376 def : Pat<(v16i16 (X86psign VR256:$src1, VR256:$src2)),
5377 (VPSIGNWrr256 VR256:$src1, VR256:$src2)>;
5378 def : Pat<(v8i32 (X86psign VR256:$src1, VR256:$src2)),
5379 (VPSIGNDrr256 VR256:$src1, VR256:$src2)>;
5381 def : Pat<(v16i16 (X86hadd VR256:$src1, VR256:$src2)),
5382 (VPHADDWrr256 VR256:$src1, VR256:$src2)>;
5383 def : Pat<(v8i32 (X86hadd VR256:$src1, VR256:$src2)),
5384 (VPHADDDrr256 VR256:$src1, VR256:$src2)>;
5385 def : Pat<(v16i16 (X86hsub VR256:$src1, VR256:$src2)),
5386 (VPHSUBWrr256 VR256:$src1, VR256:$src2)>;
5387 def : Pat<(v8i32 (X86hsub VR256:$src1, VR256:$src2)),
5388 (VPHSUBDrr256 VR256:$src1, VR256:$src2)>;
5391 //===---------------------------------------------------------------------===//
5392 // SSSE3 - Packed Align Instruction Patterns
5393 //===---------------------------------------------------------------------===//
5395 multiclass ssse3_palign<string asm, bit Is2Addr = 1> {
5396 let neverHasSideEffects = 1 in {
5397 def R128rr : SS3AI<0x0F, MRMSrcReg, (outs VR128:$dst),
5398 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
5400 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5402 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5405 def R128rm : SS3AI<0x0F, MRMSrcMem, (outs VR128:$dst),
5406 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
5408 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5410 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5415 multiclass ssse3_palign_y<string asm, bit Is2Addr = 1> {
5416 let neverHasSideEffects = 1 in {
5417 def R256rr : SS3AI<0x0F, MRMSrcReg, (outs VR256:$dst),
5418 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
5420 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5423 def R256rm : SS3AI<0x0F, MRMSrcMem, (outs VR256:$dst),
5424 (ins VR256:$src1, i256mem:$src2, i8imm:$src3),
5426 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
5431 let Predicates = [HasAVX] in
5432 defm VPALIGN : ssse3_palign<"vpalignr", 0>, VEX_4V;
5433 let Predicates = [HasAVX2] in
5434 defm VPALIGN : ssse3_palign_y<"vpalignr", 0>, VEX_4V;
5435 let Constraints = "$src1 = $dst", Predicates = [HasSSSE3] in
5436 defm PALIGN : ssse3_palign<"palignr">;
5438 let Predicates = [HasSSSE3] in {
5439 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5440 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5441 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5442 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5443 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5444 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5445 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5446 (PALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5449 let Predicates = [HasAVX] in {
5450 def : Pat<(v4i32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5451 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5452 def : Pat<(v4f32 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5453 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5454 def : Pat<(v8i16 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5455 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5456 def : Pat<(v16i8 (X86PAlign VR128:$src1, VR128:$src2, (i8 imm:$imm))),
5457 (VPALIGNR128rr VR128:$src2, VR128:$src1, imm:$imm)>;
5460 //===---------------------------------------------------------------------===//
5461 // SSSE3 - Thread synchronization
5462 //===---------------------------------------------------------------------===//
5464 let usesCustomInserter = 1 in {
5465 def MONITOR : PseudoI<(outs), (ins i32mem:$src1, GR32:$src2, GR32:$src3),
5466 [(int_x86_sse3_monitor addr:$src1, GR32:$src2, GR32:$src3)]>;
5467 def MWAIT : PseudoI<(outs), (ins GR32:$src1, GR32:$src2),
5468 [(int_x86_sse3_mwait GR32:$src1, GR32:$src2)]>;
5471 let Uses = [EAX, ECX, EDX] in
5472 def MONITORrrr : I<0x01, MRM_C8, (outs), (ins), "monitor", []>, TB,
5473 Requires<[HasSSE3]>;
5474 let Uses = [ECX, EAX] in
5475 def MWAITrr : I<0x01, MRM_C9, (outs), (ins), "mwait", []>, TB,
5476 Requires<[HasSSE3]>;
5478 def : InstAlias<"mwait %eax, %ecx", (MWAITrr)>, Requires<[In32BitMode]>;
5479 def : InstAlias<"mwait %rax, %rcx", (MWAITrr)>, Requires<[In64BitMode]>;
5481 def : InstAlias<"monitor %eax, %ecx, %edx", (MONITORrrr)>,
5482 Requires<[In32BitMode]>;
5483 def : InstAlias<"monitor %rax, %rcx, %rdx", (MONITORrrr)>,
5484 Requires<[In64BitMode]>;
5486 //===----------------------------------------------------------------------===//
5487 // SSE4.1 - Packed Move with Sign/Zero Extend
5488 //===----------------------------------------------------------------------===//
5490 multiclass SS41I_binop_rm_int8<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5491 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5492 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5493 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5495 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i64mem:$src),
5496 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5498 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5502 multiclass SS41I_binop_rm_int16_y<bits<8> opc, string OpcodeStr,
5504 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5505 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5506 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5508 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i128mem:$src),
5509 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5510 [(set VR256:$dst, (IntId (load addr:$src)))]>, OpSize;
5513 let Predicates = [HasAVX] in {
5514 defm VPMOVSXBW : SS41I_binop_rm_int8<0x20, "vpmovsxbw", int_x86_sse41_pmovsxbw>,
5516 defm VPMOVSXWD : SS41I_binop_rm_int8<0x23, "vpmovsxwd", int_x86_sse41_pmovsxwd>,
5518 defm VPMOVSXDQ : SS41I_binop_rm_int8<0x25, "vpmovsxdq", int_x86_sse41_pmovsxdq>,
5520 defm VPMOVZXBW : SS41I_binop_rm_int8<0x30, "vpmovzxbw", int_x86_sse41_pmovzxbw>,
5522 defm VPMOVZXWD : SS41I_binop_rm_int8<0x33, "vpmovzxwd", int_x86_sse41_pmovzxwd>,
5524 defm VPMOVZXDQ : SS41I_binop_rm_int8<0x35, "vpmovzxdq", int_x86_sse41_pmovzxdq>,
5528 let Predicates = [HasAVX2] in {
5529 defm VPMOVSXBW : SS41I_binop_rm_int16_y<0x20, "vpmovsxbw",
5530 int_x86_avx2_pmovsxbw>, VEX;
5531 defm VPMOVSXWD : SS41I_binop_rm_int16_y<0x23, "vpmovsxwd",
5532 int_x86_avx2_pmovsxwd>, VEX;
5533 defm VPMOVSXDQ : SS41I_binop_rm_int16_y<0x25, "vpmovsxdq",
5534 int_x86_avx2_pmovsxdq>, VEX;
5535 defm VPMOVZXBW : SS41I_binop_rm_int16_y<0x30, "vpmovzxbw",
5536 int_x86_avx2_pmovzxbw>, VEX;
5537 defm VPMOVZXWD : SS41I_binop_rm_int16_y<0x33, "vpmovzxwd",
5538 int_x86_avx2_pmovzxwd>, VEX;
5539 defm VPMOVZXDQ : SS41I_binop_rm_int16_y<0x35, "vpmovzxdq",
5540 int_x86_avx2_pmovzxdq>, VEX;
5543 defm PMOVSXBW : SS41I_binop_rm_int8<0x20, "pmovsxbw", int_x86_sse41_pmovsxbw>;
5544 defm PMOVSXWD : SS41I_binop_rm_int8<0x23, "pmovsxwd", int_x86_sse41_pmovsxwd>;
5545 defm PMOVSXDQ : SS41I_binop_rm_int8<0x25, "pmovsxdq", int_x86_sse41_pmovsxdq>;
5546 defm PMOVZXBW : SS41I_binop_rm_int8<0x30, "pmovzxbw", int_x86_sse41_pmovzxbw>;
5547 defm PMOVZXWD : SS41I_binop_rm_int8<0x33, "pmovzxwd", int_x86_sse41_pmovzxwd>;
5548 defm PMOVZXDQ : SS41I_binop_rm_int8<0x35, "pmovzxdq", int_x86_sse41_pmovzxdq>;
5550 let Predicates = [HasSSE41] in {
5551 // Common patterns involving scalar load.
5552 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5553 (PMOVSXBWrm addr:$src)>;
5554 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5555 (PMOVSXBWrm addr:$src)>;
5557 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5558 (PMOVSXWDrm addr:$src)>;
5559 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5560 (PMOVSXWDrm addr:$src)>;
5562 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5563 (PMOVSXDQrm addr:$src)>;
5564 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5565 (PMOVSXDQrm addr:$src)>;
5567 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5568 (PMOVZXBWrm addr:$src)>;
5569 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5570 (PMOVZXBWrm addr:$src)>;
5572 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5573 (PMOVZXWDrm addr:$src)>;
5574 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5575 (PMOVZXWDrm addr:$src)>;
5577 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5578 (PMOVZXDQrm addr:$src)>;
5579 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5580 (PMOVZXDQrm addr:$src)>;
5583 let Predicates = [HasAVX] in {
5584 // Common patterns involving scalar load.
5585 def : Pat<(int_x86_sse41_pmovsxbw (vzmovl_v2i64 addr:$src)),
5586 (VPMOVSXBWrm addr:$src)>;
5587 def : Pat<(int_x86_sse41_pmovsxbw (vzload_v2i64 addr:$src)),
5588 (VPMOVSXBWrm addr:$src)>;
5590 def : Pat<(int_x86_sse41_pmovsxwd (vzmovl_v2i64 addr:$src)),
5591 (VPMOVSXWDrm addr:$src)>;
5592 def : Pat<(int_x86_sse41_pmovsxwd (vzload_v2i64 addr:$src)),
5593 (VPMOVSXWDrm addr:$src)>;
5595 def : Pat<(int_x86_sse41_pmovsxdq (vzmovl_v2i64 addr:$src)),
5596 (VPMOVSXDQrm addr:$src)>;
5597 def : Pat<(int_x86_sse41_pmovsxdq (vzload_v2i64 addr:$src)),
5598 (VPMOVSXDQrm addr:$src)>;
5600 def : Pat<(int_x86_sse41_pmovzxbw (vzmovl_v2i64 addr:$src)),
5601 (VPMOVZXBWrm addr:$src)>;
5602 def : Pat<(int_x86_sse41_pmovzxbw (vzload_v2i64 addr:$src)),
5603 (VPMOVZXBWrm addr:$src)>;
5605 def : Pat<(int_x86_sse41_pmovzxwd (vzmovl_v2i64 addr:$src)),
5606 (VPMOVZXWDrm addr:$src)>;
5607 def : Pat<(int_x86_sse41_pmovzxwd (vzload_v2i64 addr:$src)),
5608 (VPMOVZXWDrm addr:$src)>;
5610 def : Pat<(int_x86_sse41_pmovzxdq (vzmovl_v2i64 addr:$src)),
5611 (VPMOVZXDQrm addr:$src)>;
5612 def : Pat<(int_x86_sse41_pmovzxdq (vzload_v2i64 addr:$src)),
5613 (VPMOVZXDQrm addr:$src)>;
5617 multiclass SS41I_binop_rm_int4<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5618 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5619 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5620 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5622 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i32mem:$src),
5623 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5625 (IntId (bitconvert (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5629 multiclass SS41I_binop_rm_int8_y<bits<8> opc, string OpcodeStr,
5631 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5632 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5633 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5635 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i32mem:$src),
5636 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5638 (IntId (bitconvert (v2i64 (scalar_to_vector (loadi64 addr:$src))))))]>,
5642 let Predicates = [HasAVX] in {
5643 defm VPMOVSXBD : SS41I_binop_rm_int4<0x21, "vpmovsxbd", int_x86_sse41_pmovsxbd>,
5645 defm VPMOVSXWQ : SS41I_binop_rm_int4<0x24, "vpmovsxwq", int_x86_sse41_pmovsxwq>,
5647 defm VPMOVZXBD : SS41I_binop_rm_int4<0x31, "vpmovzxbd", int_x86_sse41_pmovzxbd>,
5649 defm VPMOVZXWQ : SS41I_binop_rm_int4<0x34, "vpmovzxwq", int_x86_sse41_pmovzxwq>,
5653 let Predicates = [HasAVX2] in {
5654 defm VPMOVSXBD : SS41I_binop_rm_int8_y<0x21, "vpmovsxbd",
5655 int_x86_avx2_pmovsxbd>, VEX;
5656 defm VPMOVSXWQ : SS41I_binop_rm_int8_y<0x24, "vpmovsxwq",
5657 int_x86_avx2_pmovsxwq>, VEX;
5658 defm VPMOVZXBD : SS41I_binop_rm_int8_y<0x31, "vpmovzxbd",
5659 int_x86_avx2_pmovzxbd>, VEX;
5660 defm VPMOVZXWQ : SS41I_binop_rm_int8_y<0x34, "vpmovzxwq",
5661 int_x86_avx2_pmovzxwq>, VEX;
5664 defm PMOVSXBD : SS41I_binop_rm_int4<0x21, "pmovsxbd", int_x86_sse41_pmovsxbd>;
5665 defm PMOVSXWQ : SS41I_binop_rm_int4<0x24, "pmovsxwq", int_x86_sse41_pmovsxwq>;
5666 defm PMOVZXBD : SS41I_binop_rm_int4<0x31, "pmovzxbd", int_x86_sse41_pmovzxbd>;
5667 defm PMOVZXWQ : SS41I_binop_rm_int4<0x34, "pmovzxwq", int_x86_sse41_pmovzxwq>;
5669 let Predicates = [HasSSE41] in {
5670 // Common patterns involving scalar load
5671 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5672 (PMOVSXBDrm addr:$src)>;
5673 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5674 (PMOVSXWQrm addr:$src)>;
5676 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5677 (PMOVZXBDrm addr:$src)>;
5678 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5679 (PMOVZXWQrm addr:$src)>;
5682 let Predicates = [HasAVX] in {
5683 // Common patterns involving scalar load
5684 def : Pat<(int_x86_sse41_pmovsxbd (vzmovl_v4i32 addr:$src)),
5685 (VPMOVSXBDrm addr:$src)>;
5686 def : Pat<(int_x86_sse41_pmovsxwq (vzmovl_v4i32 addr:$src)),
5687 (VPMOVSXWQrm addr:$src)>;
5689 def : Pat<(int_x86_sse41_pmovzxbd (vzmovl_v4i32 addr:$src)),
5690 (VPMOVZXBDrm addr:$src)>;
5691 def : Pat<(int_x86_sse41_pmovzxwq (vzmovl_v4i32 addr:$src)),
5692 (VPMOVZXWQrm addr:$src)>;
5695 multiclass SS41I_binop_rm_int2<bits<8> opc, string OpcodeStr, Intrinsic IntId> {
5696 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
5697 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5698 [(set VR128:$dst, (IntId VR128:$src))]>, OpSize;
5700 // Expecting a i16 load any extended to i32 value.
5701 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst), (ins i16mem:$src),
5702 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5703 [(set VR128:$dst, (IntId (bitconvert
5704 (v4i32 (scalar_to_vector (loadi16_anyext addr:$src))))))]>,
5708 multiclass SS41I_binop_rm_int4_y<bits<8> opc, string OpcodeStr,
5710 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
5711 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5712 [(set VR256:$dst, (IntId VR128:$src))]>, OpSize;
5714 // Expecting a i16 load any extended to i32 value.
5715 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst), (ins i16mem:$src),
5716 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
5717 [(set VR256:$dst, (IntId (bitconvert
5718 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))]>,
5722 let Predicates = [HasAVX] in {
5723 defm VPMOVSXBQ : SS41I_binop_rm_int2<0x22, "vpmovsxbq", int_x86_sse41_pmovsxbq>,
5725 defm VPMOVZXBQ : SS41I_binop_rm_int2<0x32, "vpmovzxbq", int_x86_sse41_pmovzxbq>,
5728 let Predicates = [HasAVX2] in {
5729 defm VPMOVSXBQ : SS41I_binop_rm_int4_y<0x22, "vpmovsxbq",
5730 int_x86_avx2_pmovsxbq>, VEX;
5731 defm VPMOVZXBQ : SS41I_binop_rm_int4_y<0x32, "vpmovzxbq",
5732 int_x86_avx2_pmovzxbq>, VEX;
5734 defm PMOVSXBQ : SS41I_binop_rm_int2<0x22, "pmovsxbq", int_x86_sse41_pmovsxbq>;
5735 defm PMOVZXBQ : SS41I_binop_rm_int2<0x32, "pmovzxbq", int_x86_sse41_pmovzxbq>;
5737 let Predicates = [HasSSE41] in {
5738 // Common patterns involving scalar load
5739 def : Pat<(int_x86_sse41_pmovsxbq
5740 (bitconvert (v4i32 (X86vzmovl
5741 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5742 (PMOVSXBQrm addr:$src)>;
5744 def : Pat<(int_x86_sse41_pmovzxbq
5745 (bitconvert (v4i32 (X86vzmovl
5746 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5747 (PMOVZXBQrm addr:$src)>;
5750 let Predicates = [HasAVX] in {
5751 // Common patterns involving scalar load
5752 def : Pat<(int_x86_sse41_pmovsxbq
5753 (bitconvert (v4i32 (X86vzmovl
5754 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5755 (VPMOVSXBQrm addr:$src)>;
5757 def : Pat<(int_x86_sse41_pmovzxbq
5758 (bitconvert (v4i32 (X86vzmovl
5759 (v4i32 (scalar_to_vector (loadi32 addr:$src))))))),
5760 (VPMOVZXBQrm addr:$src)>;
5763 //===----------------------------------------------------------------------===//
5764 // SSE4.1 - Extract Instructions
5765 //===----------------------------------------------------------------------===//
5767 /// SS41I_binop_ext8 - SSE 4.1 extract 8 bits to 32 bit reg or 8 bit mem
5768 multiclass SS41I_extract8<bits<8> opc, string OpcodeStr> {
5769 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5770 (ins VR128:$src1, i32i8imm:$src2),
5771 !strconcat(OpcodeStr,
5772 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5773 [(set GR32:$dst, (X86pextrb (v16i8 VR128:$src1), imm:$src2))]>,
5775 let neverHasSideEffects = 1, mayStore = 1 in
5776 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5777 (ins i8mem:$dst, VR128:$src1, i32i8imm:$src2),
5778 !strconcat(OpcodeStr,
5779 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5782 // There's an AssertZext in the way of writing the store pattern
5783 // (store (i8 (trunc (X86pextrb (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5786 let Predicates = [HasAVX] in {
5787 defm VPEXTRB : SS41I_extract8<0x14, "vpextrb">, VEX;
5788 def VPEXTRBrr64 : SS4AIi8<0x14, MRMDestReg, (outs GR64:$dst),
5789 (ins VR128:$src1, i32i8imm:$src2),
5790 "vpextrb\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, OpSize, VEX;
5793 defm PEXTRB : SS41I_extract8<0x14, "pextrb">;
5796 /// SS41I_extract16 - SSE 4.1 extract 16 bits to memory destination
5797 multiclass SS41I_extract16<bits<8> opc, string OpcodeStr> {
5798 let neverHasSideEffects = 1, mayStore = 1 in
5799 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5800 (ins i16mem:$dst, VR128:$src1, i32i8imm:$src2),
5801 !strconcat(OpcodeStr,
5802 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5805 // There's an AssertZext in the way of writing the store pattern
5806 // (store (i16 (trunc (X86pextrw (v16i8 VR128:$src1), imm:$src2))), addr:$dst)
5809 let Predicates = [HasAVX] in
5810 defm VPEXTRW : SS41I_extract16<0x15, "vpextrw">, VEX;
5812 defm PEXTRW : SS41I_extract16<0x15, "pextrw">;
5815 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5816 multiclass SS41I_extract32<bits<8> opc, string OpcodeStr> {
5817 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5818 (ins VR128:$src1, i32i8imm:$src2),
5819 !strconcat(OpcodeStr,
5820 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5822 (extractelt (v4i32 VR128:$src1), imm:$src2))]>, OpSize;
5823 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5824 (ins i32mem:$dst, VR128:$src1, i32i8imm:$src2),
5825 !strconcat(OpcodeStr,
5826 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5827 [(store (extractelt (v4i32 VR128:$src1), imm:$src2),
5828 addr:$dst)]>, OpSize;
5831 let Predicates = [HasAVX] in
5832 defm VPEXTRD : SS41I_extract32<0x16, "vpextrd">, VEX;
5834 defm PEXTRD : SS41I_extract32<0x16, "pextrd">;
5836 /// SS41I_extract32 - SSE 4.1 extract 32 bits to int reg or memory destination
5837 multiclass SS41I_extract64<bits<8> opc, string OpcodeStr> {
5838 def rr : SS4AIi8<opc, MRMDestReg, (outs GR64:$dst),
5839 (ins VR128:$src1, i32i8imm:$src2),
5840 !strconcat(OpcodeStr,
5841 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5843 (extractelt (v2i64 VR128:$src1), imm:$src2))]>, OpSize, REX_W;
5844 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5845 (ins i64mem:$dst, VR128:$src1, i32i8imm:$src2),
5846 !strconcat(OpcodeStr,
5847 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5848 [(store (extractelt (v2i64 VR128:$src1), imm:$src2),
5849 addr:$dst)]>, OpSize, REX_W;
5852 let Predicates = [HasAVX] in
5853 defm VPEXTRQ : SS41I_extract64<0x16, "vpextrq">, VEX, VEX_W;
5855 defm PEXTRQ : SS41I_extract64<0x16, "pextrq">;
5857 /// SS41I_extractf32 - SSE 4.1 extract 32 bits fp value to int reg or memory
5859 multiclass SS41I_extractf32<bits<8> opc, string OpcodeStr> {
5860 def rr : SS4AIi8<opc, MRMDestReg, (outs GR32:$dst),
5861 (ins VR128:$src1, i32i8imm:$src2),
5862 !strconcat(OpcodeStr,
5863 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5865 (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2))]>,
5867 def mr : SS4AIi8<opc, MRMDestMem, (outs),
5868 (ins f32mem:$dst, VR128:$src1, i32i8imm:$src2),
5869 !strconcat(OpcodeStr,
5870 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
5871 [(store (extractelt (bc_v4i32 (v4f32 VR128:$src1)), imm:$src2),
5872 addr:$dst)]>, OpSize;
5875 let ExeDomain = SSEPackedSingle in {
5876 let Predicates = [HasAVX] in {
5877 defm VEXTRACTPS : SS41I_extractf32<0x17, "vextractps">, VEX;
5878 def VEXTRACTPSrr64 : SS4AIi8<0x17, MRMDestReg, (outs GR64:$dst),
5879 (ins VR128:$src1, i32i8imm:$src2),
5880 "vextractps \t{$src2, $src1, $dst|$dst, $src1, $src2}",
5883 defm EXTRACTPS : SS41I_extractf32<0x17, "extractps">;
5886 // Also match an EXTRACTPS store when the store is done as f32 instead of i32.
5887 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5890 (EXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5891 Requires<[HasSSE41]>;
5892 def : Pat<(store (f32 (bitconvert (extractelt (bc_v4i32 (v4f32 VR128:$src1)),
5895 (VEXTRACTPSmr addr:$dst, VR128:$src1, imm:$src2)>,
5898 //===----------------------------------------------------------------------===//
5899 // SSE4.1 - Insert Instructions
5900 //===----------------------------------------------------------------------===//
5902 multiclass SS41I_insert8<bits<8> opc, string asm, bit Is2Addr = 1> {
5903 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5904 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5906 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5908 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5910 (X86pinsrb VR128:$src1, GR32:$src2, imm:$src3))]>, OpSize;
5911 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5912 (ins VR128:$src1, i8mem:$src2, i32i8imm:$src3),
5914 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5916 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5918 (X86pinsrb VR128:$src1, (extloadi8 addr:$src2),
5919 imm:$src3))]>, OpSize;
5922 let Predicates = [HasAVX] in
5923 defm VPINSRB : SS41I_insert8<0x20, "vpinsrb", 0>, VEX_4V;
5924 let Constraints = "$src1 = $dst" in
5925 defm PINSRB : SS41I_insert8<0x20, "pinsrb">;
5927 multiclass SS41I_insert32<bits<8> opc, string asm, bit Is2Addr = 1> {
5928 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5929 (ins VR128:$src1, GR32:$src2, i32i8imm:$src3),
5931 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5933 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5935 (v4i32 (insertelt VR128:$src1, GR32:$src2, imm:$src3)))]>,
5937 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5938 (ins VR128:$src1, i32mem:$src2, i32i8imm:$src3),
5940 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5942 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5944 (v4i32 (insertelt VR128:$src1, (loadi32 addr:$src2),
5945 imm:$src3)))]>, OpSize;
5948 let Predicates = [HasAVX] in
5949 defm VPINSRD : SS41I_insert32<0x22, "vpinsrd", 0>, VEX_4V;
5950 let Constraints = "$src1 = $dst" in
5951 defm PINSRD : SS41I_insert32<0x22, "pinsrd">;
5953 multiclass SS41I_insert64<bits<8> opc, string asm, bit Is2Addr = 1> {
5954 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5955 (ins VR128:$src1, GR64:$src2, i32i8imm:$src3),
5957 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5959 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5961 (v2i64 (insertelt VR128:$src1, GR64:$src2, imm:$src3)))]>,
5963 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5964 (ins VR128:$src1, i64mem:$src2, i32i8imm:$src3),
5966 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5968 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5970 (v2i64 (insertelt VR128:$src1, (loadi64 addr:$src2),
5971 imm:$src3)))]>, OpSize;
5974 let Predicates = [HasAVX] in
5975 defm VPINSRQ : SS41I_insert64<0x22, "vpinsrq", 0>, VEX_4V, VEX_W;
5976 let Constraints = "$src1 = $dst" in
5977 defm PINSRQ : SS41I_insert64<0x22, "pinsrq">, REX_W;
5979 // insertps has a few different modes, there's the first two here below which
5980 // are optimized inserts that won't zero arbitrary elements in the destination
5981 // vector. The next one matches the intrinsic and could zero arbitrary elements
5982 // in the target vector.
5983 multiclass SS41I_insertf32<bits<8> opc, string asm, bit Is2Addr = 1> {
5984 def rr : SS4AIi8<opc, MRMSrcReg, (outs VR128:$dst),
5985 (ins VR128:$src1, VR128:$src2, u32u8imm:$src3),
5987 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5989 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
5991 (X86insrtps VR128:$src1, VR128:$src2, imm:$src3))]>,
5993 def rm : SS4AIi8<opc, MRMSrcMem, (outs VR128:$dst),
5994 (ins VR128:$src1, f32mem:$src2, u32u8imm:$src3),
5996 !strconcat(asm, "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
5998 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6000 (X86insrtps VR128:$src1,
6001 (v4f32 (scalar_to_vector (loadf32 addr:$src2))),
6002 imm:$src3))]>, OpSize;
6005 let ExeDomain = SSEPackedSingle in {
6006 let Constraints = "$src1 = $dst" in
6007 defm INSERTPS : SS41I_insertf32<0x21, "insertps">;
6008 let Predicates = [HasAVX] in
6009 defm VINSERTPS : SS41I_insertf32<0x21, "vinsertps", 0>, VEX_4V;
6012 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6013 (VINSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6015 def : Pat<(int_x86_sse41_insertps VR128:$src1, VR128:$src2, imm:$src3),
6016 (INSERTPSrr VR128:$src1, VR128:$src2, imm:$src3)>,
6017 Requires<[HasSSE41]>;
6019 //===----------------------------------------------------------------------===//
6020 // SSE4.1 - Round Instructions
6021 //===----------------------------------------------------------------------===//
6023 multiclass sse41_fp_unop_rm<bits<8> opcps, bits<8> opcpd, string OpcodeStr,
6024 X86MemOperand x86memop, RegisterClass RC,
6025 PatFrag mem_frag32, PatFrag mem_frag64,
6026 Intrinsic V4F32Int, Intrinsic V2F64Int> {
6027 let ExeDomain = SSEPackedSingle in {
6028 // Intrinsic operation, reg.
6029 // Vector intrinsic operation, reg
6030 def PSr : SS4AIi8<opcps, MRMSrcReg,
6031 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6032 !strconcat(OpcodeStr,
6033 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6034 [(set RC:$dst, (V4F32Int RC:$src1, imm:$src2))]>,
6037 // Vector intrinsic operation, mem
6038 def PSm : SS4AIi8<opcps, MRMSrcMem,
6039 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6040 !strconcat(OpcodeStr,
6041 "ps\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6043 (V4F32Int (mem_frag32 addr:$src1),imm:$src2))]>,
6045 } // ExeDomain = SSEPackedSingle
6047 let ExeDomain = SSEPackedDouble in {
6048 // Vector intrinsic operation, reg
6049 def PDr : SS4AIi8<opcpd, MRMSrcReg,
6050 (outs RC:$dst), (ins RC:$src1, i32i8imm:$src2),
6051 !strconcat(OpcodeStr,
6052 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6053 [(set RC:$dst, (V2F64Int RC:$src1, imm:$src2))]>,
6056 // Vector intrinsic operation, mem
6057 def PDm : SS4AIi8<opcpd, MRMSrcMem,
6058 (outs RC:$dst), (ins x86memop:$src1, i32i8imm:$src2),
6059 !strconcat(OpcodeStr,
6060 "pd\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6062 (V2F64Int (mem_frag64 addr:$src1),imm:$src2))]>,
6064 } // ExeDomain = SSEPackedDouble
6067 multiclass sse41_fp_binop_rm<bits<8> opcss, bits<8> opcsd,
6070 Intrinsic F64Int, bit Is2Addr = 1> {
6071 let ExeDomain = GenericDomain in {
6073 def SSr : SS4AIi8<opcss, MRMSrcReg,
6074 (outs FR32:$dst), (ins FR32:$src1, FR32:$src2, i32i8imm:$src3),
6076 !strconcat(OpcodeStr,
6077 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6078 !strconcat(OpcodeStr,
6079 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6082 // Intrinsic operation, reg.
6083 def SSr_Int : SS4AIi8<opcss, MRMSrcReg,
6084 (outs VR128:$dst), (ins VR128:$src1, VR128:$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, $src3}")),
6090 [(set VR128:$dst, (F32Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6093 // Intrinsic operation, mem.
6094 def SSm : SS4AIi8<opcss, MRMSrcMem,
6095 (outs VR128:$dst), (ins VR128:$src1, ssmem:$src2, i32i8imm:$src3),
6097 !strconcat(OpcodeStr,
6098 "ss\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6099 !strconcat(OpcodeStr,
6100 "ss\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6102 (F32Int VR128:$src1, sse_load_f32:$src2, imm:$src3))]>,
6106 def SDr : SS4AIi8<opcsd, MRMSrcReg,
6107 (outs FR64:$dst), (ins FR64:$src1, FR64:$src2, i32i8imm:$src3),
6109 !strconcat(OpcodeStr,
6110 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6111 !strconcat(OpcodeStr,
6112 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6115 // Intrinsic operation, reg.
6116 def SDr_Int : SS4AIi8<opcsd, MRMSrcReg,
6117 (outs VR128:$dst), (ins VR128:$src1, VR128:$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, $src3}")),
6123 [(set VR128:$dst, (F64Int VR128:$src1, VR128:$src2, imm:$src3))]>,
6126 // Intrinsic operation, mem.
6127 def SDm : SS4AIi8<opcsd, MRMSrcMem,
6128 (outs VR128:$dst), (ins VR128:$src1, sdmem:$src2, i32i8imm:$src3),
6130 !strconcat(OpcodeStr,
6131 "sd\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6132 !strconcat(OpcodeStr,
6133 "sd\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6135 (F64Int VR128:$src1, sse_load_f64:$src2, imm:$src3))]>,
6137 } // ExeDomain = GenericDomain
6140 // FP round - roundss, roundps, roundsd, roundpd
6141 let Predicates = [HasAVX] in {
6143 defm VROUND : sse41_fp_unop_rm<0x08, 0x09, "vround", f128mem, VR128,
6144 memopv4f32, memopv2f64,
6145 int_x86_sse41_round_ps,
6146 int_x86_sse41_round_pd>, VEX;
6147 defm VROUNDY : sse41_fp_unop_rm<0x08, 0x09, "vround", f256mem, VR256,
6148 memopv8f32, memopv4f64,
6149 int_x86_avx_round_ps_256,
6150 int_x86_avx_round_pd_256>, VEX;
6151 defm VROUND : sse41_fp_binop_rm<0x0A, 0x0B, "vround",
6152 int_x86_sse41_round_ss,
6153 int_x86_sse41_round_sd, 0>, VEX_4V, VEX_LIG;
6155 def : Pat<(ffloor FR32:$src),
6156 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6157 def : Pat<(f64 (ffloor FR64:$src)),
6158 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6159 def : Pat<(f32 (fnearbyint FR32:$src)),
6160 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6161 def : Pat<(f64 (fnearbyint FR64:$src)),
6162 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6163 def : Pat<(f32 (fceil FR32:$src)),
6164 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6165 def : Pat<(f64 (fceil FR64:$src)),
6166 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6167 def : Pat<(f32 (frint FR32:$src)),
6168 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6169 def : Pat<(f64 (frint FR64:$src)),
6170 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6171 def : Pat<(f32 (ftrunc FR32:$src)),
6172 (VROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6173 def : Pat<(f64 (ftrunc FR64:$src)),
6174 (VROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6177 defm ROUND : sse41_fp_unop_rm<0x08, 0x09, "round", f128mem, VR128,
6178 memopv4f32, memopv2f64,
6179 int_x86_sse41_round_ps, int_x86_sse41_round_pd>;
6180 let Constraints = "$src1 = $dst" in
6181 defm ROUND : sse41_fp_binop_rm<0x0A, 0x0B, "round",
6182 int_x86_sse41_round_ss, int_x86_sse41_round_sd>;
6184 def : Pat<(ffloor FR32:$src),
6185 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x1))>;
6186 def : Pat<(f64 (ffloor FR64:$src)),
6187 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x1))>;
6188 def : Pat<(f32 (fnearbyint FR32:$src)),
6189 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0xC))>;
6190 def : Pat<(f64 (fnearbyint FR64:$src)),
6191 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0xC))>;
6192 def : Pat<(f32 (fceil FR32:$src)),
6193 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x2))>;
6194 def : Pat<(f64 (fceil FR64:$src)),
6195 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x2))>;
6196 def : Pat<(f32 (frint FR32:$src)),
6197 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x4))>;
6198 def : Pat<(f64 (frint FR64:$src)),
6199 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x4))>;
6200 def : Pat<(f32 (ftrunc FR32:$src)),
6201 (ROUNDSSr (f32 (IMPLICIT_DEF)), FR32:$src, (i32 0x3))>;
6202 def : Pat<(f64 (ftrunc FR64:$src)),
6203 (ROUNDSDr (f64 (IMPLICIT_DEF)), FR64:$src, (i32 0x3))>;
6205 //===----------------------------------------------------------------------===//
6206 // SSE4.1 - Packed Bit Test
6207 //===----------------------------------------------------------------------===//
6209 // ptest instruction we'll lower to this in X86ISelLowering primarily from
6210 // the intel intrinsic that corresponds to this.
6211 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6212 def VPTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6213 "vptest\t{$src2, $src1|$src1, $src2}",
6214 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6216 def VPTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6217 "vptest\t{$src2, $src1|$src1, $src2}",
6218 [(set EFLAGS,(X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6221 def VPTESTYrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR256:$src1, VR256:$src2),
6222 "vptest\t{$src2, $src1|$src1, $src2}",
6223 [(set EFLAGS, (X86ptest VR256:$src1, (v4i64 VR256:$src2)))]>,
6225 def VPTESTYrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR256:$src1, i256mem:$src2),
6226 "vptest\t{$src2, $src1|$src1, $src2}",
6227 [(set EFLAGS,(X86ptest VR256:$src1, (memopv4i64 addr:$src2)))]>,
6231 let Defs = [EFLAGS] in {
6232 def PTESTrr : SS48I<0x17, MRMSrcReg, (outs), (ins VR128:$src1, VR128:$src2),
6233 "ptest\t{$src2, $src1|$src1, $src2}",
6234 [(set EFLAGS, (X86ptest VR128:$src1, (v4f32 VR128:$src2)))]>,
6236 def PTESTrm : SS48I<0x17, MRMSrcMem, (outs), (ins VR128:$src1, f128mem:$src2),
6237 "ptest\t{$src2, $src1|$src1, $src2}",
6238 [(set EFLAGS, (X86ptest VR128:$src1, (memopv4f32 addr:$src2)))]>,
6242 // The bit test instructions below are AVX only
6243 multiclass avx_bittest<bits<8> opc, string OpcodeStr, RegisterClass RC,
6244 X86MemOperand x86memop, PatFrag mem_frag, ValueType vt> {
6245 def rr : SS48I<opc, MRMSrcReg, (outs), (ins RC:$src1, RC:$src2),
6246 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6247 [(set EFLAGS, (X86testp RC:$src1, (vt RC:$src2)))]>, OpSize, VEX;
6248 def rm : SS48I<opc, MRMSrcMem, (outs), (ins RC:$src1, x86memop:$src2),
6249 !strconcat(OpcodeStr, "\t{$src2, $src1|$src1, $src2}"),
6250 [(set EFLAGS, (X86testp RC:$src1, (mem_frag addr:$src2)))]>,
6254 let Defs = [EFLAGS], Predicates = [HasAVX] in {
6255 let ExeDomain = SSEPackedSingle in {
6256 defm VTESTPS : avx_bittest<0x0E, "vtestps", VR128, f128mem, memopv4f32, v4f32>;
6257 defm VTESTPSY : avx_bittest<0x0E, "vtestps", VR256, f256mem, memopv8f32, v8f32>;
6259 let ExeDomain = SSEPackedDouble in {
6260 defm VTESTPD : avx_bittest<0x0F, "vtestpd", VR128, f128mem, memopv2f64, v2f64>;
6261 defm VTESTPDY : avx_bittest<0x0F, "vtestpd", VR256, f256mem, memopv4f64, v4f64>;
6265 //===----------------------------------------------------------------------===//
6266 // SSE4.1 - Misc Instructions
6267 //===----------------------------------------------------------------------===//
6269 let Defs = [EFLAGS], Predicates = [HasPOPCNT] in {
6270 def POPCNT16rr : I<0xB8, MRMSrcReg, (outs GR16:$dst), (ins GR16:$src),
6271 "popcnt{w}\t{$src, $dst|$dst, $src}",
6272 [(set GR16:$dst, (ctpop GR16:$src)), (implicit EFLAGS)]>,
6274 def POPCNT16rm : I<0xB8, MRMSrcMem, (outs GR16:$dst), (ins i16mem:$src),
6275 "popcnt{w}\t{$src, $dst|$dst, $src}",
6276 [(set GR16:$dst, (ctpop (loadi16 addr:$src))),
6277 (implicit EFLAGS)]>, OpSize, XS;
6279 def POPCNT32rr : I<0xB8, MRMSrcReg, (outs GR32:$dst), (ins GR32:$src),
6280 "popcnt{l}\t{$src, $dst|$dst, $src}",
6281 [(set GR32:$dst, (ctpop GR32:$src)), (implicit EFLAGS)]>,
6283 def POPCNT32rm : I<0xB8, MRMSrcMem, (outs GR32:$dst), (ins i32mem:$src),
6284 "popcnt{l}\t{$src, $dst|$dst, $src}",
6285 [(set GR32:$dst, (ctpop (loadi32 addr:$src))),
6286 (implicit EFLAGS)]>, XS;
6288 def POPCNT64rr : RI<0xB8, MRMSrcReg, (outs GR64:$dst), (ins GR64:$src),
6289 "popcnt{q}\t{$src, $dst|$dst, $src}",
6290 [(set GR64:$dst, (ctpop GR64:$src)), (implicit EFLAGS)]>,
6292 def POPCNT64rm : RI<0xB8, MRMSrcMem, (outs GR64:$dst), (ins i64mem:$src),
6293 "popcnt{q}\t{$src, $dst|$dst, $src}",
6294 [(set GR64:$dst, (ctpop (loadi64 addr:$src))),
6295 (implicit EFLAGS)]>, XS;
6300 // SS41I_unop_rm_int_v16 - SSE 4.1 unary operator whose type is v8i16.
6301 multiclass SS41I_unop_rm_int_v16<bits<8> opc, string OpcodeStr,
6302 Intrinsic IntId128> {
6303 def rr128 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6305 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6306 [(set VR128:$dst, (IntId128 VR128:$src))]>, OpSize;
6307 def rm128 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6309 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
6312 (bitconvert (memopv2i64 addr:$src))))]>, OpSize;
6315 let Predicates = [HasAVX] in
6316 defm VPHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "vphminposuw",
6317 int_x86_sse41_phminposuw>, VEX;
6318 defm PHMINPOSUW : SS41I_unop_rm_int_v16 <0x41, "phminposuw",
6319 int_x86_sse41_phminposuw>;
6321 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6322 multiclass SS41I_binop_rm_int<bits<8> opc, string OpcodeStr,
6323 Intrinsic IntId128, bit Is2Addr = 1> {
6324 let isCommutable = 1 in
6325 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6326 (ins VR128:$src1, VR128:$src2),
6328 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6329 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6330 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>, OpSize;
6331 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6332 (ins VR128:$src1, i128mem:$src2),
6334 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6335 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6337 (IntId128 VR128:$src1,
6338 (bitconvert (memopv2i64 addr:$src2))))]>, OpSize;
6341 /// SS41I_binop_rm_int - Simple SSE 4.1 binary operator
6342 multiclass SS41I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6343 Intrinsic IntId256> {
6344 let isCommutable = 1 in
6345 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6346 (ins VR256:$src1, VR256:$src2),
6347 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6348 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>, OpSize;
6349 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6350 (ins VR256:$src1, i256mem:$src2),
6351 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6353 (IntId256 VR256:$src1,
6354 (bitconvert (memopv4i64 addr:$src2))))]>, OpSize;
6357 let Predicates = [HasAVX] in {
6358 let isCommutable = 0 in
6359 defm VPACKUSDW : SS41I_binop_rm_int<0x2B, "vpackusdw", int_x86_sse41_packusdw,
6361 defm VPCMPEQQ : SS41I_binop_rm_int<0x29, "vpcmpeqq", int_x86_sse41_pcmpeqq,
6363 defm VPMINSB : SS41I_binop_rm_int<0x38, "vpminsb", int_x86_sse41_pminsb,
6365 defm VPMINSD : SS41I_binop_rm_int<0x39, "vpminsd", int_x86_sse41_pminsd,
6367 defm VPMINUD : SS41I_binop_rm_int<0x3B, "vpminud", int_x86_sse41_pminud,
6369 defm VPMINUW : SS41I_binop_rm_int<0x3A, "vpminuw", int_x86_sse41_pminuw,
6371 defm VPMAXSB : SS41I_binop_rm_int<0x3C, "vpmaxsb", int_x86_sse41_pmaxsb,
6373 defm VPMAXSD : SS41I_binop_rm_int<0x3D, "vpmaxsd", int_x86_sse41_pmaxsd,
6375 defm VPMAXUD : SS41I_binop_rm_int<0x3F, "vpmaxud", int_x86_sse41_pmaxud,
6377 defm VPMAXUW : SS41I_binop_rm_int<0x3E, "vpmaxuw", int_x86_sse41_pmaxuw,
6379 defm VPMULDQ : SS41I_binop_rm_int<0x28, "vpmuldq", int_x86_sse41_pmuldq,
6382 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6383 (VPCMPEQQrr VR128:$src1, VR128:$src2)>;
6384 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6385 (VPCMPEQQrm VR128:$src1, addr:$src2)>;
6388 let Predicates = [HasAVX2] in {
6389 let isCommutable = 0 in
6390 defm VPACKUSDW : SS41I_binop_rm_int_y<0x2B, "vpackusdw",
6391 int_x86_avx2_packusdw>, VEX_4V;
6392 defm VPCMPEQQ : SS41I_binop_rm_int_y<0x29, "vpcmpeqq",
6393 int_x86_avx2_pcmpeq_q>, VEX_4V;
6394 defm VPMINSB : SS41I_binop_rm_int_y<0x38, "vpminsb",
6395 int_x86_avx2_pmins_b>, VEX_4V;
6396 defm VPMINSD : SS41I_binop_rm_int_y<0x39, "vpminsd",
6397 int_x86_avx2_pmins_d>, VEX_4V;
6398 defm VPMINUD : SS41I_binop_rm_int_y<0x3B, "vpminud",
6399 int_x86_avx2_pminu_d>, VEX_4V;
6400 defm VPMINUW : SS41I_binop_rm_int_y<0x3A, "vpminuw",
6401 int_x86_avx2_pminu_w>, VEX_4V;
6402 defm VPMAXSB : SS41I_binop_rm_int_y<0x3C, "vpmaxsb",
6403 int_x86_avx2_pmaxs_b>, VEX_4V;
6404 defm VPMAXSD : SS41I_binop_rm_int_y<0x3D, "vpmaxsd",
6405 int_x86_avx2_pmaxs_d>, VEX_4V;
6406 defm VPMAXUD : SS41I_binop_rm_int_y<0x3F, "vpmaxud",
6407 int_x86_avx2_pmaxu_d>, VEX_4V;
6408 defm VPMAXUW : SS41I_binop_rm_int_y<0x3E, "vpmaxuw",
6409 int_x86_avx2_pmaxu_w>, VEX_4V;
6410 defm VPMULDQ : SS41I_binop_rm_int_y<0x28, "vpmuldq",
6411 int_x86_avx2_pmul_dq>, VEX_4V;
6413 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, VR256:$src2)),
6414 (VPCMPEQQYrr VR256:$src1, VR256:$src2)>;
6415 def : Pat<(v4i64 (X86pcmpeqq VR256:$src1, (memop addr:$src2))),
6416 (VPCMPEQQYrm VR256:$src1, addr:$src2)>;
6419 let Constraints = "$src1 = $dst" in {
6420 let isCommutable = 0 in
6421 defm PACKUSDW : SS41I_binop_rm_int<0x2B, "packusdw", int_x86_sse41_packusdw>;
6422 defm PCMPEQQ : SS41I_binop_rm_int<0x29, "pcmpeqq", int_x86_sse41_pcmpeqq>;
6423 defm PMINSB : SS41I_binop_rm_int<0x38, "pminsb", int_x86_sse41_pminsb>;
6424 defm PMINSD : SS41I_binop_rm_int<0x39, "pminsd", int_x86_sse41_pminsd>;
6425 defm PMINUD : SS41I_binop_rm_int<0x3B, "pminud", int_x86_sse41_pminud>;
6426 defm PMINUW : SS41I_binop_rm_int<0x3A, "pminuw", int_x86_sse41_pminuw>;
6427 defm PMAXSB : SS41I_binop_rm_int<0x3C, "pmaxsb", int_x86_sse41_pmaxsb>;
6428 defm PMAXSD : SS41I_binop_rm_int<0x3D, "pmaxsd", int_x86_sse41_pmaxsd>;
6429 defm PMAXUD : SS41I_binop_rm_int<0x3F, "pmaxud", int_x86_sse41_pmaxud>;
6430 defm PMAXUW : SS41I_binop_rm_int<0x3E, "pmaxuw", int_x86_sse41_pmaxuw>;
6431 defm PMULDQ : SS41I_binop_rm_int<0x28, "pmuldq", int_x86_sse41_pmuldq>;
6434 let Predicates = [HasSSE41] in {
6435 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, VR128:$src2)),
6436 (PCMPEQQrr VR128:$src1, VR128:$src2)>;
6437 def : Pat<(v2i64 (X86pcmpeqq VR128:$src1, (memop addr:$src2))),
6438 (PCMPEQQrm VR128:$src1, addr:$src2)>;
6441 /// SS48I_binop_rm - Simple SSE41 binary operator.
6442 multiclass SS48I_binop_rm<bits<8> opc, string OpcodeStr, SDNode OpNode,
6443 ValueType OpVT, bit Is2Addr = 1> {
6444 let isCommutable = 1 in
6445 def rr : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6446 (ins VR128:$src1, VR128:$src2),
6448 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6449 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6450 [(set VR128:$dst, (OpVT (OpNode VR128:$src1, VR128:$src2)))]>,
6452 def rm : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6453 (ins VR128:$src1, i128mem:$src2),
6455 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6456 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6457 [(set VR128:$dst, (OpNode VR128:$src1,
6458 (bc_v4i32 (memopv2i64 addr:$src2))))]>,
6462 /// SS48I_binop_rm - Simple SSE41 binary operator.
6463 multiclass SS48I_binop_rm_y<bits<8> opc, string OpcodeStr, SDNode OpNode,
6465 let isCommutable = 1 in
6466 def Yrr : SS48I<opc, MRMSrcReg, (outs VR256:$dst),
6467 (ins VR256:$src1, VR256:$src2),
6468 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6469 [(set VR256:$dst, (OpVT (OpNode VR256:$src1, VR256:$src2)))]>,
6471 def Yrm : SS48I<opc, MRMSrcMem, (outs VR256:$dst),
6472 (ins VR256:$src1, i256mem:$src2),
6473 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6474 [(set VR256:$dst, (OpNode VR256:$src1,
6475 (bc_v8i32 (memopv4i64 addr:$src2))))]>,
6479 let Predicates = [HasAVX] in
6480 defm VPMULLD : SS48I_binop_rm<0x40, "vpmulld", mul, v4i32, 0>, VEX_4V;
6481 let Predicates = [HasAVX2] in
6482 defm VPMULLD : SS48I_binop_rm_y<0x40, "vpmulld", mul, v8i32>, VEX_4V;
6483 let Constraints = "$src1 = $dst" in
6484 defm PMULLD : SS48I_binop_rm<0x40, "pmulld", mul, v4i32>;
6486 /// SS41I_binop_rmi_int - SSE 4.1 binary operator with 8-bit immediate
6487 multiclass SS41I_binop_rmi_int<bits<8> opc, string OpcodeStr,
6488 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
6489 X86MemOperand x86memop, bit Is2Addr = 1> {
6490 let isCommutable = 1 in
6491 def rri : SS4AIi8<opc, MRMSrcReg, (outs RC:$dst),
6492 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
6494 !strconcat(OpcodeStr,
6495 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6496 !strconcat(OpcodeStr,
6497 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6498 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
6500 def rmi : SS4AIi8<opc, MRMSrcMem, (outs RC:$dst),
6501 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
6503 !strconcat(OpcodeStr,
6504 "\t{$src3, $src2, $dst|$dst, $src2, $src3}"),
6505 !strconcat(OpcodeStr,
6506 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}")),
6509 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
6513 let Predicates = [HasAVX] in {
6514 let isCommutable = 0 in {
6515 let ExeDomain = SSEPackedSingle in {
6516 defm VBLENDPS : SS41I_binop_rmi_int<0x0C, "vblendps", int_x86_sse41_blendps,
6517 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6518 defm VBLENDPSY : SS41I_binop_rmi_int<0x0C, "vblendps",
6519 int_x86_avx_blend_ps_256, VR256, memopv8f32, i256mem, 0>, VEX_4V;
6521 let ExeDomain = SSEPackedDouble in {
6522 defm VBLENDPD : SS41I_binop_rmi_int<0x0D, "vblendpd", int_x86_sse41_blendpd,
6523 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6524 defm VBLENDPDY : SS41I_binop_rmi_int<0x0D, "vblendpd",
6525 int_x86_avx_blend_pd_256, VR256, memopv4f64, i256mem, 0>, VEX_4V;
6527 defm VPBLENDW : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_sse41_pblendw,
6528 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6529 defm VMPSADBW : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_sse41_mpsadbw,
6530 VR128, memopv2i64, i128mem, 0>, VEX_4V;
6532 let ExeDomain = SSEPackedSingle in
6533 defm VDPPS : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_sse41_dpps,
6534 VR128, memopv4f32, i128mem, 0>, VEX_4V;
6535 let ExeDomain = SSEPackedDouble in
6536 defm VDPPD : SS41I_binop_rmi_int<0x41, "vdppd", int_x86_sse41_dppd,
6537 VR128, memopv2f64, i128mem, 0>, VEX_4V;
6538 let ExeDomain = SSEPackedSingle in
6539 defm VDPPSY : SS41I_binop_rmi_int<0x40, "vdpps", int_x86_avx_dp_ps_256,
6540 VR256, memopv8f32, i256mem, 0>, VEX_4V;
6543 let Predicates = [HasAVX2] in {
6544 let isCommutable = 0 in {
6545 defm VPBLENDWY : SS41I_binop_rmi_int<0x0E, "vpblendw", int_x86_avx2_pblendw,
6546 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6547 defm VMPSADBWY : SS41I_binop_rmi_int<0x42, "vmpsadbw", int_x86_avx2_mpsadbw,
6548 VR256, memopv4i64, i256mem, 0>, VEX_4V;
6552 let Constraints = "$src1 = $dst" in {
6553 let isCommutable = 0 in {
6554 let ExeDomain = SSEPackedSingle in
6555 defm BLENDPS : SS41I_binop_rmi_int<0x0C, "blendps", int_x86_sse41_blendps,
6556 VR128, memopv4f32, i128mem>;
6557 let ExeDomain = SSEPackedDouble in
6558 defm BLENDPD : SS41I_binop_rmi_int<0x0D, "blendpd", int_x86_sse41_blendpd,
6559 VR128, memopv2f64, i128mem>;
6560 defm PBLENDW : SS41I_binop_rmi_int<0x0E, "pblendw", int_x86_sse41_pblendw,
6561 VR128, memopv2i64, i128mem>;
6562 defm MPSADBW : SS41I_binop_rmi_int<0x42, "mpsadbw", int_x86_sse41_mpsadbw,
6563 VR128, memopv2i64, i128mem>;
6565 let ExeDomain = SSEPackedSingle in
6566 defm DPPS : SS41I_binop_rmi_int<0x40, "dpps", int_x86_sse41_dpps,
6567 VR128, memopv4f32, i128mem>;
6568 let ExeDomain = SSEPackedDouble in
6569 defm DPPD : SS41I_binop_rmi_int<0x41, "dppd", int_x86_sse41_dppd,
6570 VR128, memopv2f64, i128mem>;
6573 /// SS41I_quaternary_int_avx - AVX SSE 4.1 with 4 operators
6574 multiclass SS41I_quaternary_int_avx<bits<8> opc, string OpcodeStr,
6575 RegisterClass RC, X86MemOperand x86memop,
6576 PatFrag mem_frag, Intrinsic IntId> {
6577 def rr : Ii8<opc, MRMSrcReg, (outs RC:$dst),
6578 (ins RC:$src1, RC:$src2, RC:$src3),
6579 !strconcat(OpcodeStr,
6580 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6581 [(set RC:$dst, (IntId RC:$src1, RC:$src2, RC:$src3))],
6582 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6584 def rm : Ii8<opc, MRMSrcMem, (outs RC:$dst),
6585 (ins RC:$src1, x86memop:$src2, RC:$src3),
6586 !strconcat(OpcodeStr,
6587 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
6589 (IntId RC:$src1, (bitconvert (mem_frag addr:$src2)),
6591 SSEPackedInt>, OpSize, TA, VEX_4V, VEX_I8IMM;
6594 let Predicates = [HasAVX] in {
6595 let ExeDomain = SSEPackedDouble in {
6596 defm VBLENDVPD : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR128, i128mem,
6597 memopv2f64, int_x86_sse41_blendvpd>;
6598 defm VBLENDVPDY : SS41I_quaternary_int_avx<0x4B, "vblendvpd", VR256, i256mem,
6599 memopv4f64, int_x86_avx_blendv_pd_256>;
6600 } // ExeDomain = SSEPackedDouble
6601 let ExeDomain = SSEPackedSingle in {
6602 defm VBLENDVPS : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR128, i128mem,
6603 memopv4f32, int_x86_sse41_blendvps>;
6604 defm VBLENDVPSY : SS41I_quaternary_int_avx<0x4A, "vblendvps", VR256, i256mem,
6605 memopv8f32, int_x86_avx_blendv_ps_256>;
6606 } // ExeDomain = SSEPackedSingle
6607 defm VPBLENDVB : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR128, i128mem,
6608 memopv2i64, int_x86_sse41_pblendvb>;
6611 let Predicates = [HasAVX2] in {
6612 defm VPBLENDVBY : SS41I_quaternary_int_avx<0x4C, "vpblendvb", VR256, i256mem,
6613 memopv4i64, int_x86_avx2_pblendvb>;
6616 let Predicates = [HasAVX] in {
6617 def : Pat<(v16i8 (vselect (v16i8 VR128:$mask), (v16i8 VR128:$src1),
6618 (v16i8 VR128:$src2))),
6619 (VPBLENDVBrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6620 def : Pat<(v4i32 (vselect (v4i32 VR128:$mask), (v4i32 VR128:$src1),
6621 (v4i32 VR128:$src2))),
6622 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6623 def : Pat<(v4f32 (vselect (v4i32 VR128:$mask), (v4f32 VR128:$src1),
6624 (v4f32 VR128:$src2))),
6625 (VBLENDVPSrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6626 def : Pat<(v2i64 (vselect (v2i64 VR128:$mask), (v2i64 VR128:$src1),
6627 (v2i64 VR128:$src2))),
6628 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6629 def : Pat<(v2f64 (vselect (v2i64 VR128:$mask), (v2f64 VR128:$src1),
6630 (v2f64 VR128:$src2))),
6631 (VBLENDVPDrr VR128:$src2, VR128:$src1, VR128:$mask)>;
6632 def : Pat<(v8i32 (vselect (v8i32 VR256:$mask), (v8i32 VR256:$src1),
6633 (v8i32 VR256:$src2))),
6634 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6635 def : Pat<(v8f32 (vselect (v8i32 VR256:$mask), (v8f32 VR256:$src1),
6636 (v8f32 VR256:$src2))),
6637 (VBLENDVPSYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6638 def : Pat<(v4i64 (vselect (v4i64 VR256:$mask), (v4i64 VR256:$src1),
6639 (v4i64 VR256:$src2))),
6640 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6641 def : Pat<(v4f64 (vselect (v4i64 VR256:$mask), (v4f64 VR256:$src1),
6642 (v4f64 VR256:$src2))),
6643 (VBLENDVPDYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6646 let Predicates = [HasAVX2] in {
6647 def : Pat<(v32i8 (vselect (v32i8 VR256:$mask), (v32i8 VR256:$src1),
6648 (v32i8 VR256:$src2))),
6649 (VPBLENDVBYrr VR256:$src2, VR256:$src1, VR256:$mask)>;
6652 /// SS41I_ternary_int - SSE 4.1 ternary operator
6653 let Uses = [XMM0], Constraints = "$src1 = $dst" in {
6654 multiclass SS41I_ternary_int<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
6656 def rr0 : SS48I<opc, MRMSrcReg, (outs VR128:$dst),
6657 (ins VR128:$src1, VR128:$src2),
6658 !strconcat(OpcodeStr,
6659 "\t{$src2, $dst|$dst, $src2}"),
6660 [(set VR128:$dst, (IntId VR128:$src1, VR128:$src2, XMM0))]>,
6663 def rm0 : SS48I<opc, MRMSrcMem, (outs VR128:$dst),
6664 (ins VR128:$src1, i128mem:$src2),
6665 !strconcat(OpcodeStr,
6666 "\t{$src2, $dst|$dst, $src2}"),
6669 (bitconvert (mem_frag addr:$src2)), XMM0))]>, OpSize;
6673 let ExeDomain = SSEPackedDouble in
6674 defm BLENDVPD : SS41I_ternary_int<0x15, "blendvpd", memopv2f64,
6675 int_x86_sse41_blendvpd>;
6676 let ExeDomain = SSEPackedSingle in
6677 defm BLENDVPS : SS41I_ternary_int<0x14, "blendvps", memopv4f32,
6678 int_x86_sse41_blendvps>;
6679 defm PBLENDVB : SS41I_ternary_int<0x10, "pblendvb", memopv2i64,
6680 int_x86_sse41_pblendvb>;
6682 let Predicates = [HasSSE41] in {
6683 def : Pat<(v16i8 (vselect (v16i8 XMM0), (v16i8 VR128:$src1),
6684 (v16i8 VR128:$src2))),
6685 (PBLENDVBrr0 VR128:$src2, VR128:$src1)>;
6686 def : Pat<(v4i32 (vselect (v4i32 XMM0), (v4i32 VR128:$src1),
6687 (v4i32 VR128:$src2))),
6688 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6689 def : Pat<(v4f32 (vselect (v4i32 XMM0), (v4f32 VR128:$src1),
6690 (v4f32 VR128:$src2))),
6691 (BLENDVPSrr0 VR128:$src2, VR128:$src1)>;
6692 def : Pat<(v2i64 (vselect (v2i64 XMM0), (v2i64 VR128:$src1),
6693 (v2i64 VR128:$src2))),
6694 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6695 def : Pat<(v2f64 (vselect (v2i64 XMM0), (v2f64 VR128:$src1),
6696 (v2f64 VR128:$src2))),
6697 (BLENDVPDrr0 VR128:$src2, VR128:$src1)>;
6700 let Predicates = [HasAVX] in
6701 def VMOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6702 "vmovntdqa\t{$src, $dst|$dst, $src}",
6703 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6705 let Predicates = [HasAVX2] in
6706 def VMOVNTDQAYrm : SS48I<0x2A, MRMSrcMem, (outs VR256:$dst), (ins i256mem:$src),
6707 "vmovntdqa\t{$src, $dst|$dst, $src}",
6708 [(set VR256:$dst, (int_x86_avx2_movntdqa addr:$src))]>,
6710 def MOVNTDQArm : SS48I<0x2A, MRMSrcMem, (outs VR128:$dst), (ins i128mem:$src),
6711 "movntdqa\t{$src, $dst|$dst, $src}",
6712 [(set VR128:$dst, (int_x86_sse41_movntdqa addr:$src))]>,
6715 //===----------------------------------------------------------------------===//
6716 // SSE4.2 - Compare Instructions
6717 //===----------------------------------------------------------------------===//
6719 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6720 multiclass SS42I_binop_rm_int<bits<8> opc, string OpcodeStr,
6721 Intrinsic IntId128, bit Is2Addr = 1> {
6722 def rr : SS428I<opc, MRMSrcReg, (outs VR128:$dst),
6723 (ins VR128:$src1, VR128:$src2),
6725 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6726 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6727 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
6729 def rm : SS428I<opc, MRMSrcMem, (outs VR128:$dst),
6730 (ins VR128:$src1, i128mem:$src2),
6732 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
6733 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
6735 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
6738 /// SS42I_binop_rm_int - Simple SSE 4.2 binary operator
6739 multiclass SS42I_binop_rm_int_y<bits<8> opc, string OpcodeStr,
6740 Intrinsic IntId256> {
6741 def Yrr : SS428I<opc, MRMSrcReg, (outs VR256:$dst),
6742 (ins VR256:$src1, VR256:$src2),
6743 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6744 [(set VR256:$dst, (IntId256 VR256:$src1, VR256:$src2))]>,
6746 def Yrm : SS428I<opc, MRMSrcMem, (outs VR256:$dst),
6747 (ins VR256:$src1, i256mem:$src2),
6748 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
6750 (IntId256 VR256:$src1, (memopv4i64 addr:$src2)))]>, OpSize;
6753 let Predicates = [HasAVX] in {
6754 defm VPCMPGTQ : SS42I_binop_rm_int<0x37, "vpcmpgtq", int_x86_sse42_pcmpgtq,
6757 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6758 (VPCMPGTQrr VR128:$src1, VR128:$src2)>;
6759 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6760 (VPCMPGTQrm VR128:$src1, addr:$src2)>;
6763 let Predicates = [HasAVX2] in {
6764 defm VPCMPGTQ : SS42I_binop_rm_int_y<0x37, "vpcmpgtq", int_x86_avx2_pcmpgt_q>,
6767 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, VR256:$src2)),
6768 (VPCMPGTQYrr VR256:$src1, VR256:$src2)>;
6769 def : Pat<(v4i64 (X86pcmpgtq VR256:$src1, (memop addr:$src2))),
6770 (VPCMPGTQYrm VR256:$src1, addr:$src2)>;
6773 let Constraints = "$src1 = $dst" in
6774 defm PCMPGTQ : SS42I_binop_rm_int<0x37, "pcmpgtq", int_x86_sse42_pcmpgtq>;
6776 let Predicates = [HasSSE42] in {
6777 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, VR128:$src2)),
6778 (PCMPGTQrr VR128:$src1, VR128:$src2)>;
6779 def : Pat<(v2i64 (X86pcmpgtq VR128:$src1, (memop addr:$src2))),
6780 (PCMPGTQrm VR128:$src1, addr:$src2)>;
6783 //===----------------------------------------------------------------------===//
6784 // SSE4.2 - String/text Processing Instructions
6785 //===----------------------------------------------------------------------===//
6787 // Packed Compare Implicit Length Strings, Return Mask
6788 multiclass pseudo_pcmpistrm<string asm> {
6789 def REG : PseudoI<(outs VR128:$dst),
6790 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6791 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128 VR128:$src1, VR128:$src2,
6793 def MEM : PseudoI<(outs VR128:$dst),
6794 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6795 [(set VR128:$dst, (int_x86_sse42_pcmpistrm128
6796 VR128:$src1, (load addr:$src2), imm:$src3))]>;
6799 let Defs = [EFLAGS], usesCustomInserter = 1 in {
6800 defm PCMPISTRM128 : pseudo_pcmpistrm<"#PCMPISTRM128">, Requires<[HasSSE42]>;
6801 defm VPCMPISTRM128 : pseudo_pcmpistrm<"#VPCMPISTRM128">, Requires<[HasAVX]>;
6804 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1, Predicates = [HasAVX] in {
6805 def VPCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6806 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6807 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6809 def VPCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6810 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6811 "vpcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize, VEX;
6814 let Defs = [XMM0, EFLAGS], neverHasSideEffects = 1 in {
6815 def PCMPISTRM128rr : SS42AI<0x62, MRMSrcReg, (outs),
6816 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6817 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6819 def PCMPISTRM128rm : SS42AI<0x62, MRMSrcMem, (outs),
6820 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6821 "pcmpistrm\t{$src3, $src2, $src1|$src1, $src2, $src3}", []>, OpSize;
6824 // Packed Compare Explicit Length Strings, Return Mask
6825 multiclass pseudo_pcmpestrm<string asm> {
6826 def REG : PseudoI<(outs VR128:$dst),
6827 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6828 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6829 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5))]>;
6830 def MEM : PseudoI<(outs VR128:$dst),
6831 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6832 [(set VR128:$dst, (int_x86_sse42_pcmpestrm128
6833 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5))]>;
6836 let Defs = [EFLAGS], Uses = [EAX, EDX], usesCustomInserter = 1 in {
6837 defm PCMPESTRM128 : pseudo_pcmpestrm<"#PCMPESTRM128">, Requires<[HasSSE42]>;
6838 defm VPCMPESTRM128 : pseudo_pcmpestrm<"#VPCMPESTRM128">, Requires<[HasAVX]>;
6841 let Predicates = [HasAVX],
6842 Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6843 def VPCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6844 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6845 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6847 def VPCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6848 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6849 "vpcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize, VEX;
6852 let Defs = [XMM0, EFLAGS], Uses = [EAX, EDX], neverHasSideEffects = 1 in {
6853 def PCMPESTRM128rr : SS42AI<0x60, MRMSrcReg, (outs),
6854 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6855 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6857 def PCMPESTRM128rm : SS42AI<0x60, MRMSrcMem, (outs),
6858 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6859 "pcmpestrm\t{$src5, $src3, $src1|$src1, $src3, $src5}", []>, OpSize;
6862 // Packed Compare Implicit Length Strings, Return Index
6863 let Defs = [ECX, EFLAGS] in {
6864 multiclass SS42AI_pcmpistri<Intrinsic IntId128, string asm = "pcmpistri"> {
6865 def rr : SS42AI<0x63, MRMSrcReg, (outs),
6866 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
6867 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6868 [(set ECX, (IntId128 VR128:$src1, VR128:$src2, imm:$src3)),
6869 (implicit EFLAGS)]>, OpSize;
6870 def rm : SS42AI<0x63, MRMSrcMem, (outs),
6871 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
6872 !strconcat(asm, "\t{$src3, $src2, $src1|$src1, $src2, $src3}"),
6873 [(set ECX, (IntId128 VR128:$src1, (load addr:$src2), imm:$src3)),
6874 (implicit EFLAGS)]>, OpSize;
6878 let Predicates = [HasAVX] in {
6879 defm VPCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128, "vpcmpistri">,
6881 defm VPCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128, "vpcmpistri">,
6883 defm VPCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128, "vpcmpistri">,
6885 defm VPCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128, "vpcmpistri">,
6887 defm VPCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128, "vpcmpistri">,
6889 defm VPCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128, "vpcmpistri">,
6893 defm PCMPISTRI : SS42AI_pcmpistri<int_x86_sse42_pcmpistri128>;
6894 defm PCMPISTRIA : SS42AI_pcmpistri<int_x86_sse42_pcmpistria128>;
6895 defm PCMPISTRIC : SS42AI_pcmpistri<int_x86_sse42_pcmpistric128>;
6896 defm PCMPISTRIO : SS42AI_pcmpistri<int_x86_sse42_pcmpistrio128>;
6897 defm PCMPISTRIS : SS42AI_pcmpistri<int_x86_sse42_pcmpistris128>;
6898 defm PCMPISTRIZ : SS42AI_pcmpistri<int_x86_sse42_pcmpistriz128>;
6900 // Packed Compare Explicit Length Strings, Return Index
6901 let Defs = [ECX, EFLAGS], Uses = [EAX, EDX] in {
6902 multiclass SS42AI_pcmpestri<Intrinsic IntId128, string asm = "pcmpestri"> {
6903 def rr : SS42AI<0x61, MRMSrcReg, (outs),
6904 (ins VR128:$src1, VR128:$src3, i8imm:$src5),
6905 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6906 [(set ECX, (IntId128 VR128:$src1, EAX, VR128:$src3, EDX, imm:$src5)),
6907 (implicit EFLAGS)]>, OpSize;
6908 def rm : SS42AI<0x61, MRMSrcMem, (outs),
6909 (ins VR128:$src1, i128mem:$src3, i8imm:$src5),
6910 !strconcat(asm, "\t{$src5, $src3, $src1|$src1, $src3, $src5}"),
6912 (IntId128 VR128:$src1, EAX, (load addr:$src3), EDX, imm:$src5)),
6913 (implicit EFLAGS)]>, OpSize;
6917 let Predicates = [HasAVX] in {
6918 defm VPCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128, "vpcmpestri">,
6920 defm VPCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128, "vpcmpestri">,
6922 defm VPCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128, "vpcmpestri">,
6924 defm VPCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128, "vpcmpestri">,
6926 defm VPCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128, "vpcmpestri">,
6928 defm VPCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128, "vpcmpestri">,
6932 defm PCMPESTRI : SS42AI_pcmpestri<int_x86_sse42_pcmpestri128>;
6933 defm PCMPESTRIA : SS42AI_pcmpestri<int_x86_sse42_pcmpestria128>;
6934 defm PCMPESTRIC : SS42AI_pcmpestri<int_x86_sse42_pcmpestric128>;
6935 defm PCMPESTRIO : SS42AI_pcmpestri<int_x86_sse42_pcmpestrio128>;
6936 defm PCMPESTRIS : SS42AI_pcmpestri<int_x86_sse42_pcmpestris128>;
6937 defm PCMPESTRIZ : SS42AI_pcmpestri<int_x86_sse42_pcmpestriz128>;
6939 //===----------------------------------------------------------------------===//
6940 // SSE4.2 - CRC Instructions
6941 //===----------------------------------------------------------------------===//
6943 // No CRC instructions have AVX equivalents
6945 // crc intrinsic instruction
6946 // This set of instructions are only rm, the only difference is the size
6948 let Constraints = "$src1 = $dst" in {
6949 def CRC32r32m8 : SS42FI<0xF0, MRMSrcMem, (outs GR32:$dst),
6950 (ins GR32:$src1, i8mem:$src2),
6951 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6953 (int_x86_sse42_crc32_32_8 GR32:$src1,
6954 (load addr:$src2)))]>;
6955 def CRC32r32r8 : SS42FI<0xF0, MRMSrcReg, (outs GR32:$dst),
6956 (ins GR32:$src1, GR8:$src2),
6957 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6959 (int_x86_sse42_crc32_32_8 GR32:$src1, GR8:$src2))]>;
6960 def CRC32r32m16 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6961 (ins GR32:$src1, i16mem:$src2),
6962 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6964 (int_x86_sse42_crc32_32_16 GR32:$src1,
6965 (load addr:$src2)))]>,
6967 def CRC32r32r16 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6968 (ins GR32:$src1, GR16:$src2),
6969 "crc32{w} \t{$src2, $src1|$src1, $src2}",
6971 (int_x86_sse42_crc32_32_16 GR32:$src1, GR16:$src2))]>,
6973 def CRC32r32m32 : SS42FI<0xF1, MRMSrcMem, (outs GR32:$dst),
6974 (ins GR32:$src1, i32mem:$src2),
6975 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6977 (int_x86_sse42_crc32_32_32 GR32:$src1,
6978 (load addr:$src2)))]>;
6979 def CRC32r32r32 : SS42FI<0xF1, MRMSrcReg, (outs GR32:$dst),
6980 (ins GR32:$src1, GR32:$src2),
6981 "crc32{l} \t{$src2, $src1|$src1, $src2}",
6983 (int_x86_sse42_crc32_32_32 GR32:$src1, GR32:$src2))]>;
6984 def CRC32r64m8 : SS42FI<0xF0, MRMSrcMem, (outs GR64:$dst),
6985 (ins GR64:$src1, i8mem:$src2),
6986 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6988 (int_x86_sse42_crc32_64_8 GR64:$src1,
6989 (load addr:$src2)))]>,
6991 def CRC32r64r8 : SS42FI<0xF0, MRMSrcReg, (outs GR64:$dst),
6992 (ins GR64:$src1, GR8:$src2),
6993 "crc32{b} \t{$src2, $src1|$src1, $src2}",
6995 (int_x86_sse42_crc32_64_8 GR64:$src1, GR8:$src2))]>,
6997 def CRC32r64m64 : SS42FI<0xF1, MRMSrcMem, (outs GR64:$dst),
6998 (ins GR64:$src1, i64mem:$src2),
6999 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7001 (int_x86_sse42_crc32_64_64 GR64:$src1,
7002 (load addr:$src2)))]>,
7004 def CRC32r64r64 : SS42FI<0xF1, MRMSrcReg, (outs GR64:$dst),
7005 (ins GR64:$src1, GR64:$src2),
7006 "crc32{q} \t{$src2, $src1|$src1, $src2}",
7008 (int_x86_sse42_crc32_64_64 GR64:$src1, GR64:$src2))]>,
7012 //===----------------------------------------------------------------------===//
7013 // AES-NI Instructions
7014 //===----------------------------------------------------------------------===//
7016 multiclass AESI_binop_rm_int<bits<8> opc, string OpcodeStr,
7017 Intrinsic IntId128, bit Is2Addr = 1> {
7018 def rr : AES8I<opc, MRMSrcReg, (outs VR128:$dst),
7019 (ins VR128:$src1, VR128:$src2),
7021 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7022 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7023 [(set VR128:$dst, (IntId128 VR128:$src1, VR128:$src2))]>,
7025 def rm : AES8I<opc, MRMSrcMem, (outs VR128:$dst),
7026 (ins VR128:$src1, i128mem:$src2),
7028 !strconcat(OpcodeStr, "\t{$src2, $dst|$dst, $src2}"),
7029 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}")),
7031 (IntId128 VR128:$src1, (memopv2i64 addr:$src2)))]>, OpSize;
7034 // Perform One Round of an AES Encryption/Decryption Flow
7035 let Predicates = [HasAVX, HasAES] in {
7036 defm VAESENC : AESI_binop_rm_int<0xDC, "vaesenc",
7037 int_x86_aesni_aesenc, 0>, VEX_4V;
7038 defm VAESENCLAST : AESI_binop_rm_int<0xDD, "vaesenclast",
7039 int_x86_aesni_aesenclast, 0>, VEX_4V;
7040 defm VAESDEC : AESI_binop_rm_int<0xDE, "vaesdec",
7041 int_x86_aesni_aesdec, 0>, VEX_4V;
7042 defm VAESDECLAST : AESI_binop_rm_int<0xDF, "vaesdeclast",
7043 int_x86_aesni_aesdeclast, 0>, VEX_4V;
7046 let Constraints = "$src1 = $dst" in {
7047 defm AESENC : AESI_binop_rm_int<0xDC, "aesenc",
7048 int_x86_aesni_aesenc>;
7049 defm AESENCLAST : AESI_binop_rm_int<0xDD, "aesenclast",
7050 int_x86_aesni_aesenclast>;
7051 defm AESDEC : AESI_binop_rm_int<0xDE, "aesdec",
7052 int_x86_aesni_aesdec>;
7053 defm AESDECLAST : AESI_binop_rm_int<0xDF, "aesdeclast",
7054 int_x86_aesni_aesdeclast>;
7057 // Perform the AES InvMixColumn Transformation
7058 let Predicates = [HasAVX, HasAES] in {
7059 def VAESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7061 "vaesimc\t{$src1, $dst|$dst, $src1}",
7063 (int_x86_aesni_aesimc VR128:$src1))]>,
7065 def VAESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7066 (ins i128mem:$src1),
7067 "vaesimc\t{$src1, $dst|$dst, $src1}",
7068 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7071 def AESIMCrr : AES8I<0xDB, MRMSrcReg, (outs VR128:$dst),
7073 "aesimc\t{$src1, $dst|$dst, $src1}",
7075 (int_x86_aesni_aesimc VR128:$src1))]>,
7077 def AESIMCrm : AES8I<0xDB, MRMSrcMem, (outs VR128:$dst),
7078 (ins i128mem:$src1),
7079 "aesimc\t{$src1, $dst|$dst, $src1}",
7080 [(set VR128:$dst, (int_x86_aesni_aesimc (memopv2i64 addr:$src1)))]>,
7083 // AES Round Key Generation Assist
7084 let Predicates = [HasAVX, HasAES] in {
7085 def VAESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7086 (ins VR128:$src1, i8imm:$src2),
7087 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7089 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7091 def VAESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7092 (ins i128mem:$src1, i8imm:$src2),
7093 "vaeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7095 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7098 def AESKEYGENASSIST128rr : AESAI<0xDF, MRMSrcReg, (outs VR128:$dst),
7099 (ins VR128:$src1, i8imm:$src2),
7100 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7102 (int_x86_aesni_aeskeygenassist VR128:$src1, imm:$src2))]>,
7104 def AESKEYGENASSIST128rm : AESAI<0xDF, MRMSrcMem, (outs VR128:$dst),
7105 (ins i128mem:$src1, i8imm:$src2),
7106 "aeskeygenassist\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7108 (int_x86_aesni_aeskeygenassist (memopv2i64 addr:$src1), imm:$src2))]>,
7111 //===----------------------------------------------------------------------===//
7112 // CLMUL Instructions
7113 //===----------------------------------------------------------------------===//
7115 // Carry-less Multiplication instructions
7116 let neverHasSideEffects = 1 in {
7117 let Constraints = "$src1 = $dst" in {
7118 def PCLMULQDQrr : CLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7119 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7120 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7124 def PCLMULQDQrm : CLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7125 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7126 "pclmulqdq\t{$src3, $src2, $dst|$dst, $src2, $src3}",
7130 // AVX carry-less Multiplication instructions
7131 def VPCLMULQDQrr : AVXCLMULIi8<0x44, MRMSrcReg, (outs VR128:$dst),
7132 (ins VR128:$src1, VR128:$src2, i8imm:$src3),
7133 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7137 def VPCLMULQDQrm : AVXCLMULIi8<0x44, MRMSrcMem, (outs VR128:$dst),
7138 (ins VR128:$src1, i128mem:$src2, i8imm:$src3),
7139 "vpclmulqdq\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7144 multiclass pclmul_alias<string asm, int immop> {
7145 def : InstAlias<!strconcat("pclmul", asm,
7146 "dq {$src, $dst|$dst, $src}"),
7147 (PCLMULQDQrr VR128:$dst, VR128:$src, immop)>;
7149 def : InstAlias<!strconcat("pclmul", asm,
7150 "dq {$src, $dst|$dst, $src}"),
7151 (PCLMULQDQrm VR128:$dst, i128mem:$src, immop)>;
7153 def : InstAlias<!strconcat("vpclmul", asm,
7154 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7155 (VPCLMULQDQrr VR128:$dst, VR128:$src1, VR128:$src2, immop)>;
7157 def : InstAlias<!strconcat("vpclmul", asm,
7158 "dq {$src2, $src1, $dst|$dst, $src1, $src2}"),
7159 (VPCLMULQDQrm VR128:$dst, VR128:$src1, i128mem:$src2, immop)>;
7161 defm : pclmul_alias<"hqhq", 0x11>;
7162 defm : pclmul_alias<"hqlq", 0x01>;
7163 defm : pclmul_alias<"lqhq", 0x10>;
7164 defm : pclmul_alias<"lqlq", 0x00>;
7166 //===----------------------------------------------------------------------===//
7168 //===----------------------------------------------------------------------===//
7170 //===----------------------------------------------------------------------===//
7171 // VBROADCAST - Load from memory and broadcast to all elements of the
7172 // destination operand
7174 class avx_broadcast<bits<8> opc, string OpcodeStr, RegisterClass RC,
7175 X86MemOperand x86memop, Intrinsic Int> :
7176 AVX8I<opc, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7177 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7178 [(set RC:$dst, (Int addr:$src))]>, VEX;
7180 // AVX2 adds register forms
7181 class avx2_broadcast_reg<bits<8> opc, string OpcodeStr, RegisterClass RC,
7183 AVX28I<opc, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7184 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7185 [(set RC:$dst, (Int VR128:$src))]>, VEX;
7187 let ExeDomain = SSEPackedSingle in {
7188 def VBROADCASTSSrm : avx_broadcast<0x18, "vbroadcastss", VR128, f32mem,
7189 int_x86_avx_vbroadcast_ss>;
7190 def VBROADCASTSSYrm : avx_broadcast<0x18, "vbroadcastss", VR256, f32mem,
7191 int_x86_avx_vbroadcast_ss_256>;
7193 let ExeDomain = SSEPackedDouble in
7194 def VBROADCASTSDrm : avx_broadcast<0x19, "vbroadcastsd", VR256, f64mem,
7195 int_x86_avx_vbroadcast_sd_256>;
7196 def VBROADCASTF128 : avx_broadcast<0x1A, "vbroadcastf128", VR256, f128mem,
7197 int_x86_avx_vbroadcastf128_pd_256>;
7199 let ExeDomain = SSEPackedSingle in {
7200 def VBROADCASTSSrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR128,
7201 int_x86_avx2_vbroadcast_ss_ps>;
7202 def VBROADCASTSSYrr : avx2_broadcast_reg<0x18, "vbroadcastss", VR256,
7203 int_x86_avx2_vbroadcast_ss_ps_256>;
7205 let ExeDomain = SSEPackedDouble in
7206 def VBROADCASTSDrr : avx2_broadcast_reg<0x19, "vbroadcastsd", VR256,
7207 int_x86_avx2_vbroadcast_sd_pd_256>;
7209 let Predicates = [HasAVX2] in
7210 def VBROADCASTI128 : avx_broadcast<0x5A, "vbroadcasti128", VR256, i128mem,
7211 int_x86_avx2_vbroadcasti128>;
7213 def : Pat<(int_x86_avx_vbroadcastf128_ps_256 addr:$src),
7214 (VBROADCASTF128 addr:$src)>;
7217 //===----------------------------------------------------------------------===//
7218 // VINSERTF128 - Insert packed floating-point values
7220 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7221 def VINSERTF128rr : AVXAIi8<0x18, MRMSrcReg, (outs VR256:$dst),
7222 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7223 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7226 def VINSERTF128rm : AVXAIi8<0x18, MRMSrcMem, (outs VR256:$dst),
7227 (ins VR256:$src1, f128mem:$src2, i8imm:$src3),
7228 "vinsertf128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7232 def : Pat<(int_x86_avx_vinsertf128_pd_256 VR256:$src1, VR128:$src2, imm:$src3),
7233 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7234 def : Pat<(int_x86_avx_vinsertf128_ps_256 VR256:$src1, VR128:$src2, imm:$src3),
7235 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7236 def : Pat<(int_x86_avx_vinsertf128_si_256 VR256:$src1, VR128:$src2, imm:$src3),
7237 (VINSERTF128rr VR256:$src1, VR128:$src2, imm:$src3)>;
7239 //===----------------------------------------------------------------------===//
7240 // VEXTRACTF128 - Extract packed floating-point values
7242 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7243 def VEXTRACTF128rr : AVXAIi8<0x19, MRMDestReg, (outs VR128:$dst),
7244 (ins VR256:$src1, i8imm:$src2),
7245 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7248 def VEXTRACTF128mr : AVXAIi8<0x19, MRMDestMem, (outs),
7249 (ins f128mem:$dst, VR256:$src1, i8imm:$src2),
7250 "vextractf128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7254 def : Pat<(int_x86_avx_vextractf128_pd_256 VR256:$src1, imm:$src2),
7255 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7256 def : Pat<(int_x86_avx_vextractf128_ps_256 VR256:$src1, imm:$src2),
7257 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7258 def : Pat<(int_x86_avx_vextractf128_si_256 VR256:$src1, imm:$src2),
7259 (VEXTRACTF128rr VR256:$src1, imm:$src2)>;
7261 //===----------------------------------------------------------------------===//
7262 // VMASKMOV - Conditional SIMD Packed Loads and Stores
7264 multiclass avx_movmask_rm<bits<8> opc_rm, bits<8> opc_mr, string OpcodeStr,
7265 Intrinsic IntLd, Intrinsic IntLd256,
7266 Intrinsic IntSt, Intrinsic IntSt256> {
7267 def rm : AVX8I<opc_rm, MRMSrcMem, (outs VR128:$dst),
7268 (ins VR128:$src1, f128mem:$src2),
7269 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7270 [(set VR128:$dst, (IntLd addr:$src2, VR128:$src1))]>,
7272 def Yrm : AVX8I<opc_rm, MRMSrcMem, (outs VR256:$dst),
7273 (ins VR256:$src1, f256mem:$src2),
7274 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7275 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>,
7277 def mr : AVX8I<opc_mr, MRMDestMem, (outs),
7278 (ins f128mem:$dst, VR128:$src1, VR128:$src2),
7279 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7280 [(IntSt addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7281 def Ymr : AVX8I<opc_mr, MRMDestMem, (outs),
7282 (ins f256mem:$dst, VR256:$src1, VR256:$src2),
7283 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7284 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7287 let ExeDomain = SSEPackedSingle in
7288 defm VMASKMOVPS : avx_movmask_rm<0x2C, 0x2E, "vmaskmovps",
7289 int_x86_avx_maskload_ps,
7290 int_x86_avx_maskload_ps_256,
7291 int_x86_avx_maskstore_ps,
7292 int_x86_avx_maskstore_ps_256>;
7293 let ExeDomain = SSEPackedDouble in
7294 defm VMASKMOVPD : avx_movmask_rm<0x2D, 0x2F, "vmaskmovpd",
7295 int_x86_avx_maskload_pd,
7296 int_x86_avx_maskload_pd_256,
7297 int_x86_avx_maskstore_pd,
7298 int_x86_avx_maskstore_pd_256>;
7300 //===----------------------------------------------------------------------===//
7301 // VPERMIL - Permute Single and Double Floating-Point Values
7303 multiclass avx_permil<bits<8> opc_rm, bits<8> opc_rmi, string OpcodeStr,
7304 RegisterClass RC, X86MemOperand x86memop_f,
7305 X86MemOperand x86memop_i, PatFrag f_frag, PatFrag i_frag,
7306 Intrinsic IntVar, Intrinsic IntImm> {
7307 def rr : AVX8I<opc_rm, MRMSrcReg, (outs RC:$dst),
7308 (ins RC:$src1, RC:$src2),
7309 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7310 [(set RC:$dst, (IntVar RC:$src1, RC:$src2))]>, VEX_4V;
7311 def rm : AVX8I<opc_rm, MRMSrcMem, (outs RC:$dst),
7312 (ins RC:$src1, x86memop_i:$src2),
7313 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7314 [(set RC:$dst, (IntVar RC:$src1,
7315 (bitconvert (i_frag addr:$src2))))]>, VEX_4V;
7317 def ri : AVXAIi8<opc_rmi, MRMSrcReg, (outs RC:$dst),
7318 (ins RC:$src1, i8imm:$src2),
7319 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7320 [(set RC:$dst, (IntImm RC:$src1, imm:$src2))]>, VEX;
7321 def mi : AVXAIi8<opc_rmi, MRMSrcMem, (outs RC:$dst),
7322 (ins x86memop_f:$src1, i8imm:$src2),
7323 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7324 [(set RC:$dst, (IntImm (f_frag addr:$src1), imm:$src2))]>, VEX;
7327 let ExeDomain = SSEPackedSingle in {
7328 defm VPERMILPS : avx_permil<0x0C, 0x04, "vpermilps", VR128, f128mem, i128mem,
7329 memopv4f32, memopv2i64,
7330 int_x86_avx_vpermilvar_ps,
7331 int_x86_avx_vpermil_ps>;
7332 defm VPERMILPSY : avx_permil<0x0C, 0x04, "vpermilps", VR256, f256mem, i256mem,
7333 memopv8f32, memopv4i64,
7334 int_x86_avx_vpermilvar_ps_256,
7335 int_x86_avx_vpermil_ps_256>;
7337 let ExeDomain = SSEPackedDouble in {
7338 defm VPERMILPD : avx_permil<0x0D, 0x05, "vpermilpd", VR128, f128mem, i128mem,
7339 memopv2f64, memopv2i64,
7340 int_x86_avx_vpermilvar_pd,
7341 int_x86_avx_vpermil_pd>;
7342 defm VPERMILPDY : avx_permil<0x0D, 0x05, "vpermilpd", VR256, f256mem, i256mem,
7343 memopv4f64, memopv4i64,
7344 int_x86_avx_vpermilvar_pd_256,
7345 int_x86_avx_vpermil_pd_256>;
7348 def : Pat<(v8f32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7349 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7350 def : Pat<(v4f64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7351 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7352 def : Pat<(v8i32 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7353 (VPERMILPSYri VR256:$src1, imm:$imm)>;
7354 def : Pat<(v4i64 (X86VPermilp VR256:$src1, (i8 imm:$imm))),
7355 (VPERMILPDYri VR256:$src1, imm:$imm)>;
7356 def : Pat<(v8f32 (X86VPermilp (memopv8f32 addr:$src1), (i8 imm:$imm))),
7357 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7358 def : Pat<(v4f64 (X86VPermilp (memopv4f64 addr:$src1), (i8 imm:$imm))),
7359 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7360 def : Pat<(v8i32 (X86VPermilp (bc_v8i32 (memopv4i64 addr:$src1)),
7362 (VPERMILPSYmi addr:$src1, imm:$imm)>;
7363 def : Pat<(v4i64 (X86VPermilp (memopv4i64 addr:$src1), (i8 imm:$imm))),
7364 (VPERMILPDYmi addr:$src1, imm:$imm)>;
7366 //===----------------------------------------------------------------------===//
7367 // VPERM2F128 - Permute Floating-Point Values in 128-bit chunks
7369 let neverHasSideEffects = 1, ExeDomain = SSEPackedSingle in {
7370 def VPERM2F128rr : AVXAIi8<0x06, MRMSrcReg, (outs VR256:$dst),
7371 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7372 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7375 def VPERM2F128rm : AVXAIi8<0x06, MRMSrcMem, (outs VR256:$dst),
7376 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7377 "vperm2f128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7381 def : Pat<(int_x86_avx_vperm2f128_ps_256 VR256:$src1, VR256:$src2, imm:$src3),
7382 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7383 def : Pat<(int_x86_avx_vperm2f128_pd_256 VR256:$src1, VR256:$src2, imm:$src3),
7384 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7385 def : Pat<(int_x86_avx_vperm2f128_si_256 VR256:$src1, VR256:$src2, imm:$src3),
7386 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$src3)>;
7388 def : Pat<(int_x86_avx_vperm2f128_ps_256
7389 VR256:$src1, (memopv8f32 addr:$src2), imm:$src3),
7390 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7391 def : Pat<(int_x86_avx_vperm2f128_pd_256
7392 VR256:$src1, (memopv4f64 addr:$src2), imm:$src3),
7393 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7394 def : Pat<(int_x86_avx_vperm2f128_si_256
7395 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)), imm:$src3),
7396 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$src3)>;
7398 //===----------------------------------------------------------------------===//
7399 // VZERO - Zero YMM registers
7401 let Defs = [YMM0, YMM1, YMM2, YMM3, YMM4, YMM5, YMM6, YMM7,
7402 YMM8, YMM9, YMM10, YMM11, YMM12, YMM13, YMM14, YMM15] in {
7403 // Zero All YMM registers
7404 def VZEROALL : I<0x77, RawFrm, (outs), (ins), "vzeroall",
7405 [(int_x86_avx_vzeroall)]>, TB, VEX, VEX_L, Requires<[HasAVX]>;
7407 // Zero Upper bits of YMM registers
7408 def VZEROUPPER : I<0x77, RawFrm, (outs), (ins), "vzeroupper",
7409 [(int_x86_avx_vzeroupper)]>, TB, VEX, Requires<[HasAVX]>;
7412 //===----------------------------------------------------------------------===//
7413 // Half precision conversion instructions
7414 //===----------------------------------------------------------------------===//
7415 multiclass f16c_ph2ps<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7416 let Predicates = [HasAVX, HasF16C] in {
7417 def rr : I<0x13, MRMSrcReg, (outs RC:$dst), (ins VR128:$src),
7418 "vcvtph2ps\t{$src, $dst|$dst, $src}",
7419 [(set RC:$dst, (Int VR128:$src))]>,
7421 let neverHasSideEffects = 1, mayLoad = 1 in
7422 def rm : I<0x13, MRMSrcMem, (outs RC:$dst), (ins x86memop:$src),
7423 "vcvtph2ps\t{$src, $dst|$dst, $src}", []>, T8, OpSize, VEX;
7427 multiclass f16c_ps2ph<RegisterClass RC, X86MemOperand x86memop, Intrinsic Int> {
7428 let Predicates = [HasAVX, HasF16C] in {
7429 def rr : Ii8<0x1D, MRMDestReg, (outs VR128:$dst),
7430 (ins RC:$src1, i32i8imm:$src2),
7431 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7432 [(set VR128:$dst, (Int RC:$src1, imm:$src2))]>,
7434 let neverHasSideEffects = 1, mayLoad = 1 in
7435 def mr : Ii8<0x1D, MRMDestMem, (outs x86memop:$dst),
7436 (ins RC:$src1, i32i8imm:$src2),
7437 "vcvtps2ph\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>,
7442 defm VCVTPH2PS : f16c_ph2ps<VR128, f64mem, int_x86_vcvtph2ps_128>;
7443 defm VCVTPH2PSY : f16c_ph2ps<VR256, f128mem, int_x86_vcvtph2ps_256>;
7444 defm VCVTPS2PH : f16c_ps2ph<VR128, f64mem, int_x86_vcvtps2ph_128>;
7445 defm VCVTPS2PHY : f16c_ps2ph<VR256, f128mem, int_x86_vcvtps2ph_256>;
7447 //===----------------------------------------------------------------------===//
7448 // AVX2 Instructions
7449 //===----------------------------------------------------------------------===//
7451 /// AVX2_binop_rmi_int - AVX2 binary operator with 8-bit immediate
7452 multiclass AVX2_binop_rmi_int<bits<8> opc, string OpcodeStr,
7453 Intrinsic IntId, RegisterClass RC, PatFrag memop_frag,
7454 X86MemOperand x86memop> {
7455 let isCommutable = 1 in
7456 def rri : AVX2AIi8<opc, MRMSrcReg, (outs RC:$dst),
7457 (ins RC:$src1, RC:$src2, u32u8imm:$src3),
7458 !strconcat(OpcodeStr,
7459 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7460 [(set RC:$dst, (IntId RC:$src1, RC:$src2, imm:$src3))]>,
7462 def rmi : AVX2AIi8<opc, MRMSrcMem, (outs RC:$dst),
7463 (ins RC:$src1, x86memop:$src2, u32u8imm:$src3),
7464 !strconcat(OpcodeStr,
7465 "\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}"),
7468 (bitconvert (memop_frag addr:$src2)), imm:$src3))]>,
7472 let isCommutable = 0 in {
7473 defm VPBLENDD : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_128,
7474 VR128, memopv2i64, i128mem>;
7475 defm VPBLENDDY : AVX2_binop_rmi_int<0x02, "vpblendd", int_x86_avx2_pblendd_256,
7476 VR256, memopv4i64, i256mem>;
7479 //===----------------------------------------------------------------------===//
7480 // VPBROADCAST - Load from memory and broadcast to all elements of the
7481 // destination operand
7483 multiclass avx2_broadcast<bits<8> opc, string OpcodeStr,
7484 X86MemOperand x86memop, PatFrag ld_frag,
7485 Intrinsic Int128, Intrinsic Int256> {
7486 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst), (ins VR128:$src),
7487 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7488 [(set VR128:$dst, (Int128 VR128:$src))]>, VEX;
7489 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst), (ins x86memop:$src),
7490 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7492 (Int128 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7493 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst), (ins VR128:$src),
7494 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7495 [(set VR256:$dst, (Int256 VR128:$src))]>, VEX;
7496 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst), (ins x86memop:$src),
7497 !strconcat(OpcodeStr, "\t{$src, $dst|$dst, $src}"),
7499 (Int256 (scalar_to_vector (ld_frag addr:$src))))]>, VEX;
7502 defm VPBROADCASTB : avx2_broadcast<0x78, "vpbroadcastb", i8mem, loadi8,
7503 int_x86_avx2_pbroadcastb_128,
7504 int_x86_avx2_pbroadcastb_256>;
7505 defm VPBROADCASTW : avx2_broadcast<0x79, "vpbroadcastw", i16mem, loadi16,
7506 int_x86_avx2_pbroadcastw_128,
7507 int_x86_avx2_pbroadcastw_256>;
7508 defm VPBROADCASTD : avx2_broadcast<0x58, "vpbroadcastd", i32mem, loadi32,
7509 int_x86_avx2_pbroadcastd_128,
7510 int_x86_avx2_pbroadcastd_256>;
7511 defm VPBROADCASTQ : avx2_broadcast<0x59, "vpbroadcastq", i64mem, loadi64,
7512 int_x86_avx2_pbroadcastq_128,
7513 int_x86_avx2_pbroadcastq_256>;
7515 let Predicates = [HasAVX2] in {
7516 def : Pat<(v16i8 (X86VBroadcast (loadi8 addr:$src))),
7517 (VPBROADCASTBrm addr:$src)>;
7518 def : Pat<(v32i8 (X86VBroadcast (loadi8 addr:$src))),
7519 (VPBROADCASTBYrm addr:$src)>;
7520 def : Pat<(v8i16 (X86VBroadcast (loadi16 addr:$src))),
7521 (VPBROADCASTWrm addr:$src)>;
7522 def : Pat<(v16i16 (X86VBroadcast (loadi16 addr:$src))),
7523 (VPBROADCASTWYrm addr:$src)>;
7524 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7525 (VPBROADCASTDrm addr:$src)>;
7526 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7527 (VPBROADCASTDYrm addr:$src)>;
7528 def : Pat<(v2i64 (X86VBroadcast (loadi64 addr:$src))),
7529 (VPBROADCASTQrm addr:$src)>;
7530 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7531 (VPBROADCASTQYrm addr:$src)>;
7534 // AVX1 broadcast patterns
7535 def : Pat<(v8i32 (X86VBroadcast (loadi32 addr:$src))),
7536 (VBROADCASTSSYrm addr:$src)>;
7537 def : Pat<(v4i64 (X86VBroadcast (loadi64 addr:$src))),
7538 (VBROADCASTSDrm addr:$src)>;
7539 def : Pat<(v8f32 (X86VBroadcast (loadf32 addr:$src))),
7540 (VBROADCASTSSYrm addr:$src)>;
7541 def : Pat<(v4f64 (X86VBroadcast (loadf64 addr:$src))),
7542 (VBROADCASTSDrm addr:$src)>;
7544 def : Pat<(v4f32 (X86VBroadcast (loadf32 addr:$src))),
7545 (VBROADCASTSSrm addr:$src)>;
7546 def : Pat<(v4i32 (X86VBroadcast (loadi32 addr:$src))),
7547 (VBROADCASTSSrm addr:$src)>;
7549 //===----------------------------------------------------------------------===//
7550 // VPERM - Permute instructions
7553 multiclass avx2_perm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7555 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7556 (ins VR256:$src1, VR256:$src2),
7557 !strconcat(OpcodeStr,
7558 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7559 [(set VR256:$dst, (Int VR256:$src1, VR256:$src2))]>, VEX_4V;
7560 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7561 (ins VR256:$src1, i256mem:$src2),
7562 !strconcat(OpcodeStr,
7563 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7564 [(set VR256:$dst, (Int VR256:$src1,
7565 (bitconvert (mem_frag addr:$src2))))]>,
7569 defm VPERMD : avx2_perm<0x36, "vpermd", memopv4i64, int_x86_avx2_permd>;
7570 let ExeDomain = SSEPackedSingle in
7571 defm VPERMPS : avx2_perm<0x16, "vpermps", memopv8f32, int_x86_avx2_permps>;
7573 multiclass avx2_perm_imm<bits<8> opc, string OpcodeStr, PatFrag mem_frag,
7575 def Yrr : AVX2AIi8<opc, MRMSrcReg, (outs VR256:$dst),
7576 (ins VR256:$src1, i8imm:$src2),
7577 !strconcat(OpcodeStr,
7578 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7579 [(set VR256:$dst, (Int VR256:$src1, imm:$src2))]>, VEX;
7580 def Yrm : AVX2AIi8<opc, MRMSrcMem, (outs VR256:$dst),
7581 (ins i256mem:$src1, i8imm:$src2),
7582 !strconcat(OpcodeStr,
7583 "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7584 [(set VR256:$dst, (Int (mem_frag addr:$src1), imm:$src2))]>,
7588 defm VPERMQ : avx2_perm_imm<0x00, "vpermq", memopv4i64, int_x86_avx2_permq>,
7590 let ExeDomain = SSEPackedDouble in
7591 defm VPERMPD : avx2_perm_imm<0x01, "vpermpd", memopv4f64, int_x86_avx2_permpd>,
7594 //===----------------------------------------------------------------------===//
7595 // VPERM2I128 - Permute Floating-Point Values in 128-bit chunks
7597 def VPERM2I128rr : AVX2AIi8<0x46, MRMSrcReg, (outs VR256:$dst),
7598 (ins VR256:$src1, VR256:$src2, i8imm:$src3),
7599 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7601 (int_x86_avx2_vperm2i128 VR256:$src1, VR256:$src2, imm:$src3))]>,
7603 def VPERM2I128rm : AVX2AIi8<0x46, MRMSrcMem, (outs VR256:$dst),
7604 (ins VR256:$src1, f256mem:$src2, i8imm:$src3),
7605 "vperm2i128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7607 (int_x86_avx2_vperm2i128 VR256:$src1, (memopv4i64 addr:$src2),
7611 let Predicates = [HasAVX2] in {
7612 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7613 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7614 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7615 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7616 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7617 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7618 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7619 (VPERM2I128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7621 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, (bc_v32i8 (memopv4i64 addr:$src2)),
7623 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7624 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7625 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7626 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7627 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, (bc_v8i32 (memopv4i64 addr:$src2)),
7629 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7630 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, (memopv4i64 addr:$src2),
7632 (VPERM2I128rm VR256:$src1, addr:$src2, imm:$imm)>;
7636 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7637 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7638 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7639 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7640 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7641 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7642 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7643 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7644 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7645 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7646 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1, VR256:$src2, (i8 imm:$imm))),
7647 (VPERM2F128rr VR256:$src1, VR256:$src2, imm:$imm)>;
7649 def : Pat<(v8f32 (X86VPerm2x128 VR256:$src1,
7650 (memopv8f32 addr:$src2), (i8 imm:$imm))),
7651 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7652 def : Pat<(v8i32 (X86VPerm2x128 VR256:$src1,
7653 (bc_v8i32 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7654 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7655 def : Pat<(v4i64 (X86VPerm2x128 VR256:$src1,
7656 (memopv4i64 addr:$src2), (i8 imm:$imm))),
7657 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7658 def : Pat<(v4f64 (X86VPerm2x128 VR256:$src1,
7659 (memopv4f64 addr:$src2), (i8 imm:$imm))),
7660 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7661 def : Pat<(v32i8 (X86VPerm2x128 VR256:$src1,
7662 (bc_v32i8 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7663 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7664 def : Pat<(v16i16 (X86VPerm2x128 VR256:$src1,
7665 (bc_v16i16 (memopv4i64 addr:$src2)), (i8 imm:$imm))),
7666 (VPERM2F128rm VR256:$src1, addr:$src2, imm:$imm)>;
7669 //===----------------------------------------------------------------------===//
7670 // VINSERTI128 - Insert packed integer values
7672 def VINSERTI128rr : AVX2AIi8<0x38, MRMSrcReg, (outs VR256:$dst),
7673 (ins VR256:$src1, VR128:$src2, i8imm:$src3),
7674 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7676 (int_x86_avx2_vinserti128 VR256:$src1, VR128:$src2, imm:$src3))]>,
7678 def VINSERTI128rm : AVX2AIi8<0x38, MRMSrcMem, (outs VR256:$dst),
7679 (ins VR256:$src1, i128mem:$src2, i8imm:$src3),
7680 "vinserti128\t{$src3, $src2, $src1, $dst|$dst, $src1, $src2, $src3}",
7682 (int_x86_avx2_vinserti128 VR256:$src1, (memopv2i64 addr:$src2),
7683 imm:$src3))]>, VEX_4V;
7685 let Predicates = [HasAVX2] in {
7686 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7688 (VINSERTI128rr VR256:$src1, VR128:$src2,
7689 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7690 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7692 (VINSERTI128rr VR256:$src1, VR128:$src2,
7693 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7694 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7696 (VINSERTI128rr VR256:$src1, VR128:$src2,
7697 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7698 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7700 (VINSERTI128rr VR256:$src1, VR128:$src2,
7701 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7705 def : Pat<(vinsertf128_insert:$ins (v8f32 VR256:$src1), (v4f32 VR128:$src2),
7707 (VINSERTF128rr VR256:$src1, VR128:$src2,
7708 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7709 def : Pat<(vinsertf128_insert:$ins (v4f64 VR256:$src1), (v2f64 VR128:$src2),
7711 (VINSERTF128rr VR256:$src1, VR128:$src2,
7712 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7713 def : Pat<(vinsertf128_insert:$ins (v4i64 VR256:$src1), (v2i64 VR128:$src2),
7715 (VINSERTF128rr VR256:$src1, VR128:$src2,
7716 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7717 def : Pat<(vinsertf128_insert:$ins (v8i32 VR256:$src1), (v4i32 VR128:$src2),
7719 (VINSERTF128rr VR256:$src1, VR128:$src2,
7720 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7721 def : Pat<(vinsertf128_insert:$ins (v32i8 VR256:$src1), (v16i8 VR128:$src2),
7723 (VINSERTF128rr VR256:$src1, VR128:$src2,
7724 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7725 def : Pat<(vinsertf128_insert:$ins (v16i16 VR256:$src1), (v8i16 VR128:$src2),
7727 (VINSERTF128rr VR256:$src1, VR128:$src2,
7728 (INSERT_get_vinsertf128_imm VR256:$ins))>;
7730 //===----------------------------------------------------------------------===//
7731 // VEXTRACTI128 - Extract packed integer values
7733 def VEXTRACTI128rr : AVX2AIi8<0x39, MRMDestReg, (outs VR128:$dst),
7734 (ins VR256:$src1, i8imm:$src2),
7735 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}",
7737 (int_x86_avx2_vextracti128 VR256:$src1, imm:$src2))]>,
7739 let neverHasSideEffects = 1, mayStore = 1 in
7740 def VEXTRACTI128mr : AVX2AIi8<0x39, MRMDestMem, (outs),
7741 (ins i128mem:$dst, VR256:$src1, i8imm:$src2),
7742 "vextracti128\t{$src2, $src1, $dst|$dst, $src1, $src2}", []>, VEX;
7744 let Predicates = [HasAVX2] in {
7745 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7746 (v2i64 (VEXTRACTI128rr
7747 (v4i64 VR256:$src1),
7748 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7749 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7750 (v4i32 (VEXTRACTI128rr
7751 (v8i32 VR256:$src1),
7752 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7753 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7754 (v8i16 (VEXTRACTI128rr
7755 (v16i16 VR256:$src1),
7756 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7757 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7758 (v16i8 (VEXTRACTI128rr
7759 (v32i8 VR256:$src1),
7760 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7764 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7765 (v4f32 (VEXTRACTF128rr
7766 (v8f32 VR256:$src1),
7767 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7768 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7769 (v2f64 (VEXTRACTF128rr
7770 (v4f64 VR256:$src1),
7771 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7772 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7773 (v2i64 (VEXTRACTF128rr
7774 (v4i64 VR256:$src1),
7775 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7776 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7777 (v4i32 (VEXTRACTF128rr
7778 (v8i32 VR256:$src1),
7779 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7780 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7781 (v8i16 (VEXTRACTF128rr
7782 (v16i16 VR256:$src1),
7783 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7784 def : Pat<(vextractf128_extract:$ext VR256:$src1, (i32 imm)),
7785 (v16i8 (VEXTRACTF128rr
7786 (v32i8 VR256:$src1),
7787 (EXTRACT_get_vextractf128_imm VR128:$ext)))>;
7789 //===----------------------------------------------------------------------===//
7790 // VPMASKMOV - Conditional SIMD Integer Packed Loads and Stores
7792 multiclass avx2_pmovmask<string OpcodeStr,
7793 Intrinsic IntLd128, Intrinsic IntLd256,
7794 Intrinsic IntSt128, Intrinsic IntSt256> {
7795 def rm : AVX28I<0x8c, MRMSrcMem, (outs VR128:$dst),
7796 (ins VR128:$src1, i128mem:$src2),
7797 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7798 [(set VR128:$dst, (IntLd128 addr:$src2, VR128:$src1))]>, VEX_4V;
7799 def Yrm : AVX28I<0x8c, MRMSrcMem, (outs VR256:$dst),
7800 (ins VR256:$src1, i256mem:$src2),
7801 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7802 [(set VR256:$dst, (IntLd256 addr:$src2, VR256:$src1))]>, VEX_4V;
7803 def mr : AVX28I<0x8e, MRMDestMem, (outs),
7804 (ins i128mem:$dst, VR128:$src1, VR128:$src2),
7805 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7806 [(IntSt128 addr:$dst, VR128:$src1, VR128:$src2)]>, VEX_4V;
7807 def Ymr : AVX28I<0x8e, MRMDestMem, (outs),
7808 (ins i256mem:$dst, VR256:$src1, VR256:$src2),
7809 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7810 [(IntSt256 addr:$dst, VR256:$src1, VR256:$src2)]>, VEX_4V;
7813 defm VPMASKMOVD : avx2_pmovmask<"vpmaskmovd",
7814 int_x86_avx2_maskload_d,
7815 int_x86_avx2_maskload_d_256,
7816 int_x86_avx2_maskstore_d,
7817 int_x86_avx2_maskstore_d_256>;
7818 defm VPMASKMOVQ : avx2_pmovmask<"vpmaskmovq",
7819 int_x86_avx2_maskload_q,
7820 int_x86_avx2_maskload_q_256,
7821 int_x86_avx2_maskstore_q,
7822 int_x86_avx2_maskstore_q_256>, VEX_W;
7825 //===----------------------------------------------------------------------===//
7826 // Variable Bit Shifts
7828 multiclass avx2_var_shift<bits<8> opc, string OpcodeStr, SDNode OpNode,
7829 ValueType vt128, ValueType vt256> {
7830 def rr : AVX28I<opc, MRMSrcReg, (outs VR128:$dst),
7831 (ins VR128:$src1, VR128:$src2),
7832 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7834 (vt128 (OpNode VR128:$src1, (vt128 VR128:$src2))))]>,
7836 def rm : AVX28I<opc, MRMSrcMem, (outs VR128:$dst),
7837 (ins VR128:$src1, i128mem:$src2),
7838 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7840 (vt128 (OpNode VR128:$src1,
7841 (vt128 (bitconvert (memopv2i64 addr:$src2))))))]>,
7843 def Yrr : AVX28I<opc, MRMSrcReg, (outs VR256:$dst),
7844 (ins VR256:$src1, VR256:$src2),
7845 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7847 (vt256 (OpNode VR256:$src1, (vt256 VR256:$src2))))]>,
7849 def Yrm : AVX28I<opc, MRMSrcMem, (outs VR256:$dst),
7850 (ins VR256:$src1, i256mem:$src2),
7851 !strconcat(OpcodeStr, "\t{$src2, $src1, $dst|$dst, $src1, $src2}"),
7853 (vt256 (OpNode VR256:$src1,
7854 (vt256 (bitconvert (memopv4i64 addr:$src2))))))]>,
7858 defm VPSLLVD : avx2_var_shift<0x47, "vpsllvd", shl, v4i32, v8i32>;
7859 defm VPSLLVQ : avx2_var_shift<0x47, "vpsllvq", shl, v2i64, v4i64>, VEX_W;
7860 defm VPSRLVD : avx2_var_shift<0x45, "vpsrlvd", srl, v4i32, v8i32>;
7861 defm VPSRLVQ : avx2_var_shift<0x45, "vpsrlvq", srl, v2i64, v4i64>, VEX_W;
7862 defm VPSRAVD : avx2_var_shift<0x46, "vpsravd", sra, v4i32, v8i32>;